From e6215e60319499253fccb709ef50c77d9f7fae24 Mon Sep 17 00:00:00 2001
From: Nicolas <nicolas.korjahn@uni-potsdam.de>
Date: Sun, 18 Jan 2026 18:30:21 +0100
Subject: [PATCH] Implementation of the shampoo optimizer including tests and
 experiments

---
 scripts/nn/optim/shampoo.dml                  | 499 ++++++++++++++++++
 .../shampoo_optimizer/diagram_creation.ipynb  | 461 ++++++++++++++++
 .../diagrams/diagram_metrics_mnist.png        | Bin 0 -> 65314 bytes
 .../metrics/metrics_adam_cifar.csv            |  90 ++++
 .../metrics/metrics_adam_cifar.csv.mtd        |  12 +
 .../metrics/metrics_adam_mnist.csv            |  60 +++
 .../metrics/metrics_adam_mnist.csv.mtd        |  12 +
 .../metrics/metrics_shampoo_cifar.csv         |  90 ++++
 .../metrics/metrics_shampoo_cifar.csv.mtd     |  12 +
 .../metrics_shampoo_diagonal_mnist.csv        |  60 +++
 .../metrics_shampoo_diagonal_mnist.csv.mtd    |  12 +
 .../metrics_shampoo_heuristic_cifar.csv       |  90 ++++
 .../metrics_shampoo_heuristic_cifar.csv.mtd   |  12 +
 ...trics_shampoo_heuristic_diagonal_mnist.csv |  60 +++
 ...s_shampoo_heuristic_diagonal_mnist.csv.mtd |  12 +
 .../metrics_shampoo_heuristic_mnist.csv       |  60 +++
 .../metrics_shampoo_heuristic_mnist.csv.mtd   |  12 +
 .../metrics/metrics_shampoo_mnist.csv         |  60 +++
 .../metrics/metrics_shampoo_mnist.csv.mtd     |  12 +
 .../metrics_shampoo_momentum_cifar.csv        |  90 ++++
 .../metrics_shampoo_momentum_cifar.csv.mtd    |  12 +
 ...etrics_shampoo_momentum_diagonal_mnist.csv |  60 +++
 ...cs_shampoo_momentum_diagonal_mnist.csv.mtd |  12 +
 .../metrics_shampoo_momentum_mnist.csv        |  60 +++
 .../metrics_shampoo_momentum_mnist.csv.mtd    |  12 +
 .../shampoo_optimizer_experiments.dml         | 482 +++++++++++++++++
 .../nn/component/shampoo_test.dml             | 437 +++++++++++++++
 .../applications/nn/component/shampoo_test.py | 225 ++++++++
 .../nn/component/shampoo_test2.dml            |  34 ++
 29 files changed, 3050 insertions(+)
 create mode 100644 scripts/nn/optim/shampoo.dml
 create mode 100644 scripts/staging/shampoo_optimizer/diagram_creation.ipynb
 create mode 100644 scripts/staging/shampoo_optimizer/diagrams/diagram_metrics_mnist.png
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv
 create mode 100644 scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv.mtd
 create mode 100644 scripts/staging/shampoo_optimizer/shampoo_optimizer_experiments.dml
 create mode 100644 src/test/scripts/applications/nn/component/shampoo_test.dml
 create mode 100644 src/test/scripts/applications/nn/component/shampoo_test.py
 create mode 100644 src/test/scripts/applications/nn/component/shampoo_test2.dml

diff --git a/scripts/nn/optim/shampoo.dml b/scripts/nn/optim/shampoo.dml
new file mode 100644
index 00000000000..e8832bdaf9a
--- /dev/null
+++ b/scripts/nn/optim/shampoo.dml
@@ -0,0 +1,499 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+/*
+ * Shampoo optimizer.
+ *
+ * Implementation of the Shampoo optimizer as described in:
+ *
+ *  Gupta et al., "Shampoo: Preconditioned Stochastic Tensor Optimization"
+ *  https://arxiv.org/abs/1802.09568
+ *
+ * Shampoo is a second-order optimization method that preconditions
+ * gradients using estimates of the row- and column-wise covariance
+ * of the gradients. Compared to first-order optimizers (SGD, Adam),
+ * Shampoo can converge faster but is significantly more memory-intensive.
+ *
+ * This implementation supports:
+ *  - Full-matrix Shampoo (exact preconditioning)
+ *  - Diagonal Shampoo (memory-efficient approximation)
+ *
+ * The choice between the two modes is determined by the shape of X
+ * and the preconditioner initialization.
+ */
+
+update = function(matrix[double] X, matrix[double] dX, double lr,
+                  matrix[double] preconL, matrix[double] preconR, boolean useDiag)
+    return(matrix[double] X, matrix[double] preconL, matrix[double] preconR){
+  /*
+   * Performs one optimization step using the Shampoo update rule.
+   *
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be updated (n × m)
+   *  - dX: Gradient of the loss w.r.t. X (n × m)
+   *  - lr: Learning rate.
+   *  - preconL: Left (row) preconditioner
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconR: Right (column) preconditioner
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   *
+   * Outputs:
+   *  - X: Updated parameter matrix (n × m)
+   *  - preconL: Updated left preconditioner
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconR: Updated right preconditioner
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   */    
+
+  # Full-matrix Shampoo:
+  # Only used if both dimensions are small enough
+  if(useDiag==FALSE){
+
+    preconL = preconL + dX %*% t(dX)
+    preconR = preconR + t(dX) %*% dX
+
+    [LEigenvalue, LEigenvector] = eigen(preconL)
+    preconLInvPowerRoot = LEigenvector %*% diag(LEigenvalue^(-0.25)) %*% t(LEigenvector)
+
+    [REigenvalue, REigenvector] = eigen(preconR)
+    preconRInvPowerRoot = REigenvector %*% diag(REigenvalue^(-0.25)) %*% t(REigenvector)
+
+    X = X - lr * preconLInvPowerRoot %*% dX %*% preconRInvPowerRoot
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+  } else{
+    n = nrow(dX)
+    m = ncol(dX)
+
+    preconL = preconL + rowSums(dX^2)          
+    preconR = preconR + colSums(dX^2)  
+
+    preconLScale = preconL^(-0.25)        
+    preconRScale = preconR^(-0.25)
+
+    preconLMatrix = preconLScale %*% matrix(1, rows=1, cols=m)
+    preconRMatrix = matrix(1, rows=n, cols=1) %*% preconRScale
+
+    scaledGrad = dX * preconLMatrix;
+    scaledGrad = scaledGrad * preconRMatrix;
+
+    X = X - lr * scaledGrad;
+
+  }
+}
+
+init = function(matrix[double] X, double epsilon, int useDiagThreshold)
+    return (matrix[double] preconL, matrix[double] preconR, boolean useDiag) {
+  /*
+   * Initializes the Shampoo preconditioners for a given parameter matrix.
+   *
+   * Depending on the size of X, this function initializes either:
+   *  - Full identity matrices (exact Shampoo), or
+   *  - Diagonal vectors (approximate Shampoo)
+   *
+   * This threshold is crucial to avoid excessive memory usage,
+   * as full Shampoo requires O(n^2 + m^2) memory per parameter matrix.
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be optimized (n, m)
+   *  - epsilon: Numerical stability constant
+   *  - useDiagThreshold: Dimension threshold above which diagonal
+   *                      preconditioning is used
+   *
+   * Outputs:
+   *  - preconL: Initial left preconditioner
+   *      - Full: (n × n) identity scaled by epsilon
+   *      - Diagonal: (n × 1) filled with epsilon
+   *  - preconR: Initial right preconditioner
+   *      - Full: (m × m) identity scaled by epsilon
+   *      - Diagonal: (1 × m) filled with epsilon
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   */
+
+  # Use diagonal Shampoo if parameter matrix is too large
+  if((nrow(X) > useDiagThreshold) | (ncol(X) > useDiagThreshold)){
+    preconL = matrix(epsilon, rows=nrow(X), cols=1);
+    preconR = matrix(epsilon, rows=1, cols=ncol(X));
+    useDiag = TRUE
+
+  # Use full Shampoo if parameter matrix is small enough
+  } else {
+    preconL = matrix(0, rows=nrow(X), cols=nrow(X));
+    index = 1;
+      while (index <= nrow(X)){
+          preconL[index, index] = epsilon * 1
+          index = index + 1
+      }
+    preconR = matrix(0, rows=ncol(X), cols=ncol(X));
+    index = 1;
+      while (index <= ncol(X)){
+          preconR[index, index] = epsilon * 1
+          index = index + 1
+      }
+    useDiag = FALSE
+  }
+}
+
+update_momentum = function(matrix[double] X, matrix[double] dX, double lr,
+                  matrix[double] preconL, matrix[double] preconR, 
+                  matrix[double] momentum, boolean useDiag)
+    return(matrix[double] X, matrix[double] preconL, matrix[double] preconR, 
+           matrix[double] momentum){
+  /*
+   * Performs one optimization step using the Shampoo update rule, while using momentum.
+   *
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be updated (n × m)
+   *  - dX: Gradient of the loss w.r.t. X (n × m)
+   *  - lr: Learning rate.
+   *  - preconL: Left (row) preconditioner
+   *               - Full: (n × n)
+   *               - Diagonal: (n × 1)
+   *  - preconR: Right (column) preconditioner
+   *               - Full: (m × m)
+   *               - Diagonal: (1 × m)
+   *  - momentum: momentum (n × m)
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   *
+   * Outputs:
+   *  - X: Updated parameter matrix (n × m)
+   *  - preconL: Updated left preconditioner
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconR: Updated right preconditioner
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - momentum: Updated momentum (n × m)
+   */    
+
+  # calculating the updated momentum
+  momentum = 0.9 * momentum + (0.1)*dX
+
+  # Full-matrix Shampoo:
+  # Only used if both dimensions are small enough
+  if(useDiag==FALSE){
+
+    preconL = preconL + dX %*% t(dX)
+    preconR = preconR + t(dX) %*% dX
+
+    [LEigenvalue, LEigenvector] = eigen(preconL)
+    preconLInvPowerRoot = LEigenvector %*% diag(LEigenvalue^(-0.25)) %*% t(LEigenvector)
+
+    [REigenvalue, REigenvector] = eigen(preconR)
+    preconRInvPowerRoot = REigenvector %*% diag(REigenvalue^(-0.25)) %*% t(REigenvector)
+
+    X = X - lr * preconLInvPowerRoot %*% momentum %*% preconRInvPowerRoot
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+  } else{
+      n = nrow(dX)
+      m = ncol(dX)
+
+      preconL = preconL + rowSums(dX ^ 2)
+      preconR = preconR + colSums(dX ^ 2)
+
+      preconLScale = preconL^(-0.25)        
+      preconRScale = preconR^(-0.25)
+
+      preconLMatrix = preconLScale %*% matrix(1, rows=1, cols=m)
+      preconRMatrix = matrix(1, rows=n, cols=1) %*% preconRScale
+
+      scaledGrad = momentum * preconLMatrix
+      scaledGrad = scaledGrad * preconRMatrix
+
+      X = X - lr * scaledGrad         
+    }
+}
+
+init_momentum = function(matrix[double] X, double epsilon, int useDiagThreshold)
+    return (matrix[double] preconL, matrix[double] preconR, 
+            matrix[double] momentum, boolean useDiag) {
+  /*
+   * Initializes the Shampoo preconditioners and momentum for a given parameter matrix.
+   *
+   * Depending on the size of X, this function initializes either:
+   *  - Full identity matrices (exact Shampoo), or
+   *  - Diagonal vectors (approximate Shampoo)
+   *
+   * This threshold is crucial to avoid excessive memory usage,
+   * as full Shampoo requires O(n² + m²) memory per parameter matrix.
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be optimized (n, m)
+   *  - epsilon: Numerical stability constant
+   *  - useDiagThreshold: Dimension threshold above which diagonal
+   *                      preconditioning is used
+   *
+   * Outputs:
+   *  - preconL: Initial left preconditioner
+   *      - Full: (n × n) identity scaled by epsilon
+   *      - Diagonal: (n × 1) filled with epsilon
+   *  - preconR: Initial right preconditioner
+   *      - Full: (m × m) identity scaled by epsilon
+   *      - Diagonal: (1 × m) filled with epsilon
+   *  - momentum: Initial momentum (n × m), initialized to zeros
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   */
+
+  # Use diagonal Shampoo if parameter matrix is too large
+  if((nrow(X) > useDiagThreshold) | (ncol(X) > useDiagThreshold)){
+    preconL = matrix(epsilon, rows=nrow(X), cols=1);
+    preconR = matrix(epsilon, rows=1, cols=ncol(X));
+    useDiag = TRUE
+
+  # Use full Shampoo if parameter matrix is small enough
+  } else {
+    preconL = matrix(0, rows=nrow(X), cols=nrow(X));
+    index = 1;
+      while (index <= nrow(X)){
+          preconL[index, index] = epsilon * 1
+          index = index + 1
+      }
+    preconR = matrix(0, rows=ncol(X), cols=ncol(X));
+    index = 1;
+      while (index <= ncol(X)){
+          preconR[index, index] = epsilon * 1
+          index = index + 1
+      }
+    useDiag = FALSE
+  }
+  momentum = X * 0
+}
+
+update_heuristic = function(matrix[double] X, matrix[double] dX, double lr,
+                  matrix[double] preconL, matrix[double] preconR, matrix[double] momentum, 
+                  int stepCounter, int rootEvery, int preconEvery, matrix[double] bufferL, 
+                  matrix[double] bufferR, matrix[double] preconLInvPowerRoot, 
+                  matrix[double] preconRInvPowerRoot, boolean useDiag)
+    return (matrix[double] X, matrix[double] preconL, matrix[double] preconR, 
+            matrix[double] momentum, int stepCounter, matrix[double] bufferL, 
+            matrix[double] bufferR, matrix[double] preconLInvPowerRoot, 
+            matrix[double] preconRInvPowerRoot){
+  /*
+   * Performs one optimization step using the Shampoo update rule, while using momentum 
+   * and a heuristic for runtime improvements.
+   *
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be updated (n × m)
+   *  - dX: Gradient of the loss w.r.t. X (n × m)
+   *  - lr: Learning rate.
+   *  - preconL: Left (row) preconditioner
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconR: Right (column) preconditioner
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - momentum: momentum (n × m)
+   *  - stepCounter: Step counter (int), incremented each call
+   *  - rootEvery: Frequency for recomputing inverse roots (int)
+   *  - preconEvery: Frequency for applying buffered updates to preconditioners (int)
+   *  - bufferL: Buffer accumulating left curvature updates
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - bufferR: Buffer accumulating right curvature updates
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - preconLInvPowerRoot: Cached preconL^{-1/4}
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconRInvPowerRoot: Cached preconR^{-1/4}
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   *
+   * Outputs:
+   *  - X: Updated parameter matrix (n × m)
+   *  - preconL: Updated left preconditioner 
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconR: Updated right preconditioner
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - momentum: momentum (n × m)
+   *  - stepCounter: Updated step counter (int)
+   *  - bufferL: Updated bufferL (reset to 0 when applied)
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - bufferR: Updated bufferR (reset to 0 when applied)
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   *  - preconLInvPowerRoot: Updated cached inverse root (when recomputed)
+   *      - Full: (n × n)
+   *      - Diagonal: (n × 1)
+   *  - preconRInvPowerRoot: Updated cached inverse root (when recomputed)
+   *      - Full: (m × m)
+   *      - Diagonal: (1 × m)
+   */    
+
+  # calculating the updated momentum
+  momentum = 0.9 * momentum + (0.1)*dX
+
+  # Full-matrix Shampoo:
+  # Only used if both dimensions are small enough
+  if(useDiag==FALSE){
+    bufferL = bufferL + (dX %*% t(dX))
+    bufferR = bufferR + (t(dX) %*% dX)
+
+    if ((stepCounter > 0) & (stepCounter %% preconEvery == 0)){
+      preconL = preconL + bufferL
+      preconR = preconR + bufferR
+      bufferL = bufferL * 0
+      bufferR = bufferR * 0
+    }
+
+
+    if ((stepCounter > 0) & (stepCounter %% rootEvery == 0)){
+      [LEigenvalue, LEigenvector] = eigen(preconL)
+      preconLInvPowerRoot = LEigenvector %*% diag(LEigenvalue^(-0.25)) %*% t(LEigenvector)
+
+      [REigenvalue, REigenvector] = eigen(preconR)
+      preconRInvPowerRoot = REigenvector %*% diag(REigenvalue^(-0.25)) %*% t(REigenvector)
+    }
+
+    X = X - lr * preconLInvPowerRoot %*% momentum %*% preconRInvPowerRoot
+
+    stepCounter = stepCounter + 1
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+  } else{
+    n = nrow(dX)
+    m = ncol(dX)
+
+    bufferL = bufferL + rowSums(dX ^ 2)
+    bufferR = bufferR + colSums(dX ^ 2)
+
+    if ((stepCounter > 0) & (stepCounter %% preconEvery == 0)){
+      preconL = preconL + bufferL
+      preconR = preconR + bufferR
+      bufferL = bufferL * 0
+      bufferR = bufferR * 0
+    }
+
+    if ((stepCounter > 0) & (stepCounter %% rootEvery == 0)){
+      preconLInvPowerRoot = (preconL^(-0.25))
+      preconRInvPowerRoot = (preconR^(-0.25))
+    }
+    preconLMatrix = preconLInvPowerRoot %*% matrix(1, rows=1, cols=m)
+    preconRMatrix = matrix(1, rows=n, cols=1) %*% preconRInvPowerRoot
+
+    scaledGrad = momentum * preconLMatrix
+    scaledGrad = scaledGrad * preconRMatrix
+
+    X = X - lr * scaledGrad
+    stepCounter = stepCounter + 1
+  }
+}
+
+
+
+init_heuristic = function(matrix[double] X, double epsilon, int useDiagThreshold)
+    return (matrix[double] preconL, matrix[double] preconR, int stepCounter, 
+            matrix[double] bufferL, matrix[double] bufferR, matrix[double] momentum, 
+            matrix[double] preconLInvPowerRoot, matrix[double] preconRInvPowerRoot, 
+            boolean useDiag) {
+  /*
+   * Initializes Shampoo preconditioners, buffers, cached inverse roots, 
+   * and momentum for the heuristic variant.
+   *
+   * Depending on the size of X, this function initializes either:
+   *  - Full identity matrices (exact Shampoo), or
+   *  - Diagonal vectors (approximate Shampoo)
+   *
+   * This threshold is crucial to avoid excessive memory usage,
+   * as full Shampoo requires O(n^2 + m^2) memory per parameter matrix.
+   *
+   * Inputs:
+   *  - X: Parameter matrix to be optimized (n, m)
+   *  - epsilon: Numerical stability constant
+   *  - useDiagThreshold: Dimension threshold above which diagonal
+   *                      preconditioning is used
+   *
+   * Outputs:
+   *  - preconL: Initial left preconditioner (n × n) or (n × 1)
+   *  - preconR: Initial right preconditioner (m × m) or (1 × m)
+   *  - stepCounter: Initialized to 0
+   *  - bufferL: Initialized to zeros, same shape as preconL
+   *  - bufferR: Initialized to zeros, same shape as preconR
+   *  - momentum: Initialized to zeros, same shape as X (n × m)
+   *  - preconLInvPowerRoot: Cached inverse fourth root of preconL
+   *      - Full: initialized to epsilon^{-1/4} * I (n × n)
+   *      - Diagonal: initialized to preconL^{-1/4} (n × 1)
+   *  - preconRInvPowerRoot: Cached inverse fourth root of preconR
+   *      - Full: initialized to epsilon^{-1/4} * I (m × m)
+   *      - Diagonal: initialized to preconR^{-1/4} (1 × m)
+   *  - useDiag: Boolean flag indicating whether diagonal Shampoo is used
+   */
+
+  # Use diagonal Shampoo if parameter matrix is too large
+  if((nrow(X) > useDiagThreshold) | (ncol(X) > useDiagThreshold)){
+    preconL = matrix(epsilon, rows=nrow(X), cols=1);
+    preconR = matrix(epsilon, rows=1, cols=ncol(X));
+    preconLInvPowerRoot = preconL^(-0.25)
+    preconRInvPowerRoot = preconR^(-0.25)
+    useDiag = TRUE
+
+  # Use full Shampoo if parameter matrix is small enough
+  } else {
+    preconL = matrix(0, rows=nrow(X), cols=nrow(X));
+    index = 1;
+      while (index <= nrow(X)){
+          preconL[index, index] = epsilon * 1
+          index = index + 1
+      }
+    preconR = matrix(0, rows=ncol(X), cols=ncol(X));
+    index = 1;
+      while (index <= ncol(X)){
+          preconR[index, index] = epsilon * 1
+          index = index + 1
+      }
+
+    preconLInvPowerRoot = preconL
+    i = 1
+    while(i <= nrow(preconLInvPowerRoot)) {
+      preconLInvPowerRoot[i,i] = epsilon^(-0.25)
+      i = i + 1
+    }
+
+    preconRInvPowerRoot = preconR
+    j = 1
+    while(j <= nrow(preconRInvPowerRoot)) {
+      preconRInvPowerRoot[j,j] = epsilon^(-0.25)
+      j = j + 1
+    }  
+
+    useDiag = FALSE
+  }
+  bufferR = preconR * 0
+  bufferL = preconL * 0
+  stepCounter = 0
+  momentum = X * 0
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/diagram_creation.ipynb b/scripts/staging/shampoo_optimizer/diagram_creation.ipynb
new file mode 100644
index 00000000000..251f7e6a4f3
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/diagram_creation.ipynb
@@ -0,0 +1,461 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "400c10c6",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "    epochs  train_losses  train_accuracies  val_losses  val_accuracies  \\\n",
+      "0      1.0      1.529923          0.615625    0.416351           0.873   \n",
+      "1      2.0      0.250601          0.927000    0.229524           0.916   \n",
+      "2      3.0      0.154800          0.954250    0.162009           0.940   \n",
+      "3      4.0      0.113356          0.964625    0.135608           0.955   \n",
+      "4      5.0      0.090430          0.971625    0.123288           0.963   \n",
+      "5      6.0      0.076361          0.976625    0.114561           0.968   \n",
+      "6      7.0      0.066914          0.979625    0.109109           0.969   \n",
+      "7      8.0      0.059269          0.982500    0.105038           0.971   \n",
+      "8      9.0      0.053242          0.983750    0.101710           0.971   \n",
+      "9     10.0      0.048538          0.985125    0.099977           0.971   \n",
+      "10    11.0      0.044405          0.985750    0.098380           0.971   \n",
+      "11    12.0      0.040600          0.987250    0.097781           0.972   \n",
+      "12    13.0      0.037342          0.988250    0.097132           0.974   \n",
+      "13    14.0      0.034315          0.988750    0.096922           0.974   \n",
+      "14    15.0      0.031616          0.989750    0.097324           0.976   \n",
+      "15    16.0      0.029328          0.990125    0.096786           0.974   \n",
+      "16    17.0      0.027317          0.990875    0.096020           0.975   \n",
+      "17    18.0      0.025400          0.991250    0.094932           0.975   \n",
+      "18    19.0      0.023850          0.992375    0.094315           0.975   \n",
+      "19    20.0      0.022360          0.992875    0.094448           0.975   \n",
+      "20    21.0      0.021041          0.993375    0.094068           0.976   \n",
+      "21    22.0      0.019816          0.993750    0.093970           0.976   \n",
+      "22    23.0      0.018703          0.994375    0.093682           0.976   \n",
+      "23    24.0      0.017558          0.995000    0.093167           0.976   \n",
+      "24    25.0      0.016504          0.995375    0.093128           0.976   \n",
+      "25    26.0      0.015547          0.996000    0.093137           0.976   \n",
+      "26    27.0      0.014641          0.996250    0.092622           0.975   \n",
+      "27    28.0      0.013850          0.996375    0.092436           0.975   \n",
+      "28    29.0      0.013045          0.996750    0.091737           0.974   \n",
+      "29    30.0      0.012308          0.996875    0.091197           0.974   \n",
+      "30    31.0      0.011605          0.997375    0.090616           0.973   \n",
+      "31    32.0      0.010950          0.997625    0.089798           0.973   \n",
+      "32    33.0      0.010404          0.997625    0.089589           0.973   \n",
+      "33    34.0      0.009935          0.997750    0.089403           0.973   \n",
+      "34    35.0      0.009448          0.997750    0.088787           0.973   \n",
+      "35    36.0      0.008966          0.997875    0.088232           0.974   \n",
+      "36    37.0      0.008522          0.998250    0.088025           0.973   \n",
+      "37    38.0      0.008150          0.998375    0.088028           0.973   \n",
+      "38    39.0      0.007784          0.998500    0.087388           0.973   \n",
+      "39    40.0      0.007437          0.998500    0.087394           0.973   \n",
+      "40    41.0      0.007119          0.998625    0.087162           0.973   \n",
+      "41    42.0      0.006813          0.998625    0.086621           0.974   \n",
+      "42    43.0      0.006518          0.998750    0.086615           0.975   \n",
+      "43    44.0      0.006219          0.999000    0.086452           0.973   \n",
+      "44    45.0      0.005936          0.999250    0.086182           0.973   \n",
+      "45    46.0      0.005689          0.999250    0.085994           0.973   \n",
+      "46    47.0      0.005452          0.999250    0.086067           0.973   \n",
+      "47    48.0      0.005212          0.999250    0.085938           0.973   \n",
+      "48    49.0      0.004984          0.999250    0.085857           0.973   \n",
+      "49    50.0      0.004777          0.999375    0.085784           0.973   \n",
+      "50    51.0      0.004596          0.999375    0.086051           0.973   \n",
+      "51    52.0      0.004398          0.999375    0.085878           0.974   \n",
+      "52    53.0      0.004203          0.999500    0.085930           0.974   \n",
+      "53    54.0      0.004026          0.999500    0.085631           0.974   \n",
+      "54    55.0      0.003847          0.999625    0.085347           0.975   \n",
+      "55    56.0      0.003690          0.999750    0.085356           0.975   \n",
+      "56    57.0      0.003528          0.999875    0.085206           0.975   \n",
+      "57    58.0      0.003375          1.000000    0.085179           0.976   \n",
+      "58    59.0      0.003247          1.000000    0.085019           0.976   \n",
+      "59    60.0      0.003124          1.000000    0.084813           0.977   \n",
+      "\n",
+      "   optimizer  \n",
+      "0    shampoo  \n",
+      "1    shampoo  \n",
+      "2    shampoo  \n",
+      "3    shampoo  \n",
+      "4    shampoo  \n",
+      "5    shampoo  \n",
+      "6    shampoo  \n",
+      "7    shampoo  \n",
+      "8    shampoo  \n",
+      "9    shampoo  \n",
+      "10   shampoo  \n",
+      "11   shampoo  \n",
+      "12   shampoo  \n",
+      "13   shampoo  \n",
+      "14   shampoo  \n",
+      "15   shampoo  \n",
+      "16   shampoo  \n",
+      "17   shampoo  \n",
+      "18   shampoo  \n",
+      "19   shampoo  \n",
+      "20   shampoo  \n",
+      "21   shampoo  \n",
+      "22   shampoo  \n",
+      "23   shampoo  \n",
+      "24   shampoo  \n",
+      "25   shampoo  \n",
+      "26   shampoo  \n",
+      "27   shampoo  \n",
+      "28   shampoo  \n",
+      "29   shampoo  \n",
+      "30   shampoo  \n",
+      "31   shampoo  \n",
+      "32   shampoo  \n",
+      "33   shampoo  \n",
+      "34   shampoo  \n",
+      "35   shampoo  \n",
+      "36   shampoo  \n",
+      "37   shampoo  \n",
+      "38   shampoo  \n",
+      "39   shampoo  \n",
+      "40   shampoo  \n",
+      "41   shampoo  \n",
+      "42   shampoo  \n",
+      "43   shampoo  \n",
+      "44   shampoo  \n",
+      "45   shampoo  \n",
+      "46   shampoo  \n",
+      "47   shampoo  \n",
+      "48   shampoo  \n",
+      "49   shampoo  \n",
+      "50   shampoo  \n",
+      "51   shampoo  \n",
+      "52   shampoo  \n",
+      "53   shampoo  \n",
+      "54   shampoo  \n",
+      "55   shampoo  \n",
+      "56   shampoo  \n",
+      "57   shampoo  \n",
+      "58   shampoo  \n",
+      "59   shampoo  \n",
+      "    epochs  train_losses  train_accuracies  val_losses  val_accuracies  \\\n",
+      "0      1.0      0.639533          0.801500    0.327239           0.902   \n",
+      "1      2.0      0.184752          0.945375    0.221243           0.937   \n",
+      "2      3.0      0.134242          0.959750    0.188511           0.941   \n",
+      "3      4.0      0.100303          0.970375    0.166702           0.951   \n",
+      "4      5.0      0.079379          0.976000    0.156010           0.957   \n",
+      "5      6.0      0.069202          0.979125    0.164982           0.957   \n",
+      "6      7.0      0.060592          0.981250    0.171788           0.958   \n",
+      "7      8.0      0.054129          0.982750    0.163366           0.962   \n",
+      "8      9.0      0.048112          0.985375    0.149752           0.963   \n",
+      "9     10.0      0.042714          0.987625    0.135282           0.964   \n",
+      "10    11.0      0.037989          0.989000    0.122572           0.966   \n",
+      "11    12.0      0.033369          0.990125    0.114374           0.968   \n",
+      "12    13.0      0.029285          0.991125    0.106545           0.973   \n",
+      "13    14.0      0.025636          0.992250    0.100020           0.971   \n",
+      "14    15.0      0.022314          0.993375    0.093786           0.970   \n",
+      "15    16.0      0.019570          0.994750    0.088490           0.973   \n",
+      "16    17.0      0.017252          0.996125    0.084034           0.976   \n",
+      "17    18.0      0.015548          0.996250    0.082103           0.976   \n",
+      "18    19.0      0.014139          0.996875    0.080609           0.976   \n",
+      "19    20.0      0.012960          0.997750    0.078689           0.976   \n",
+      "20    21.0      0.012004          0.997500    0.077240           0.977   \n",
+      "21    22.0      0.011114          0.997500    0.076127           0.977   \n",
+      "22    23.0      0.010281          0.998000    0.075846           0.976   \n",
+      "23    24.0      0.009678          0.998000    0.074256           0.976   \n",
+      "24    25.0      0.008997          0.998000    0.073674           0.977   \n",
+      "25    26.0      0.008524          0.998000    0.072962           0.977   \n",
+      "26    27.0      0.007974          0.998250    0.072336           0.976   \n",
+      "27    28.0      0.007519          0.998625    0.071772           0.975   \n",
+      "28    29.0      0.007067          0.998750    0.070253           0.974   \n",
+      "29    30.0      0.006566          0.999000    0.069656           0.974   \n",
+      "30    31.0      0.006089          0.999000    0.068200           0.977   \n",
+      "31    32.0      0.005608          0.999125    0.066946           0.977   \n",
+      "32    33.0      0.005141          0.999250    0.065950           0.978   \n",
+      "33    34.0      0.004759          0.999250    0.064440           0.978   \n",
+      "34    35.0      0.004381          0.999375    0.063538           0.978   \n",
+      "35    36.0      0.004047          0.999750    0.062533           0.979   \n",
+      "36    37.0      0.003787          0.999750    0.061590           0.979   \n",
+      "37    38.0      0.003497          1.000000    0.060816           0.979   \n",
+      "38    39.0      0.003270          1.000000    0.060142           0.979   \n",
+      "39    40.0      0.003096          1.000000    0.059449           0.979   \n",
+      "40    41.0      0.002915          1.000000    0.059109           0.979   \n",
+      "41    42.0      0.002747          1.000000    0.058858           0.980   \n",
+      "42    43.0      0.002602          1.000000    0.058758           0.980   \n",
+      "43    44.0      0.002453          1.000000    0.058751           0.980   \n",
+      "44    45.0      0.002325          1.000000    0.058833           0.980   \n",
+      "45    46.0      0.002200          1.000000    0.059313           0.980   \n",
+      "46    47.0      0.002075          1.000000    0.059344           0.980   \n",
+      "47    48.0      0.001972          1.000000    0.059808           0.980   \n",
+      "48    49.0      0.001866          1.000000    0.060284           0.980   \n",
+      "49    50.0      0.001785          1.000000    0.060728           0.980   \n",
+      "50    51.0      0.001703          1.000000    0.061150           0.980   \n",
+      "51    52.0      0.001626          1.000000    0.061556           0.979   \n",
+      "52    53.0      0.001560          1.000000    0.061923           0.979   \n",
+      "53    54.0      0.001492          1.000000    0.062277           0.979   \n",
+      "54    55.0      0.001441          1.000000    0.062653           0.979   \n",
+      "55    56.0      0.001381          1.000000    0.063042           0.979   \n",
+      "56    57.0      0.001333          1.000000    0.063426           0.979   \n",
+      "57    58.0      0.001282          1.000000    0.063787           0.979   \n",
+      "58    59.0      0.001239          1.000000    0.064086           0.979   \n",
+      "59    60.0      0.001195          1.000000    0.064313           0.980   \n",
+      "\n",
+      "           optimizer  \n",
+      "0   shampoo_momentum  \n",
+      "1   shampoo_momentum  \n",
+      "2   shampoo_momentum  \n",
+      "3   shampoo_momentum  \n",
+      "4   shampoo_momentum  \n",
+      "5   shampoo_momentum  \n",
+      "6   shampoo_momentum  \n",
+      "7   shampoo_momentum  \n",
+      "8   shampoo_momentum  \n",
+      "9   shampoo_momentum  \n",
+      "10  shampoo_momentum  \n",
+      "11  shampoo_momentum  \n",
+      "12  shampoo_momentum  \n",
+      "13  shampoo_momentum  \n",
+      "14  shampoo_momentum  \n",
+      "15  shampoo_momentum  \n",
+      "16  shampoo_momentum  \n",
+      "17  shampoo_momentum  \n",
+      "18  shampoo_momentum  \n",
+      "19  shampoo_momentum  \n",
+      "20  shampoo_momentum  \n",
+      "21  shampoo_momentum  \n",
+      "22  shampoo_momentum  \n",
+      "23  shampoo_momentum  \n",
+      "24  shampoo_momentum  \n",
+      "25  shampoo_momentum  \n",
+      "26  shampoo_momentum  \n",
+      "27  shampoo_momentum  \n",
+      "28  shampoo_momentum  \n",
+      "29  shampoo_momentum  \n",
+      "30  shampoo_momentum  \n",
+      "31  shampoo_momentum  \n",
+      "32  shampoo_momentum  \n",
+      "33  shampoo_momentum  \n",
+      "34  shampoo_momentum  \n",
+      "35  shampoo_momentum  \n",
+      "36  shampoo_momentum  \n",
+      "37  shampoo_momentum  \n",
+      "38  shampoo_momentum  \n",
+      "39  shampoo_momentum  \n",
+      "40  shampoo_momentum  \n",
+      "41  shampoo_momentum  \n",
+      "42  shampoo_momentum  \n",
+      "43  shampoo_momentum  \n",
+      "44  shampoo_momentum  \n",
+      "45  shampoo_momentum  \n",
+      "46  shampoo_momentum  \n",
+      "47  shampoo_momentum  \n",
+      "48  shampoo_momentum  \n",
+      "49  shampoo_momentum  \n",
+      "50  shampoo_momentum  \n",
+      "51  shampoo_momentum  \n",
+      "52  shampoo_momentum  \n",
+      "53  shampoo_momentum  \n",
+      "54  shampoo_momentum  \n",
+      "55  shampoo_momentum  \n",
+      "56  shampoo_momentum  \n",
+      "57  shampoo_momentum  \n",
+      "58  shampoo_momentum  \n",
+      "59  shampoo_momentum  \n",
+      "    epochs  train_losses  train_accuracies  val_losses  val_accuracies  \\\n",
+      "0      1.0      0.660083          0.791992    0.458462           0.844   \n",
+      "1      2.0      0.256014          0.923000    0.306052           0.909   \n",
+      "2      3.0      0.179122          0.947125    0.213270           0.933   \n",
+      "3      4.0      0.139948          0.957750    0.173651           0.942   \n",
+      "4      5.0      0.107507          0.967500    0.150592           0.952   \n",
+      "5      6.0      0.090241          0.971875    0.152722           0.949   \n",
+      "6      7.0      0.081093          0.974000    0.182535           0.943   \n",
+      "7      8.0      0.073433          0.976500    0.196313           0.936   \n",
+      "8      9.0      0.064927          0.979250    0.161399           0.948   \n",
+      "9     10.0      0.055647          0.982500    0.158130           0.949   \n",
+      "10    11.0      0.048851          0.984625    0.136853           0.958   \n",
+      "11    12.0      0.043453          0.986000    0.129564           0.963   \n",
+      "12    13.0      0.037899          0.988125    0.128100           0.964   \n",
+      "13    14.0      0.034403          0.989750    0.120332           0.967   \n",
+      "14    15.0      0.031140          0.990500    0.144659           0.956   \n",
+      "15    16.0      0.027824          0.992000    0.139351           0.959   \n",
+      "16    17.0      0.025904          0.992875    0.138057           0.960   \n",
+      "17    18.0      0.023049          0.994000    0.139043           0.958   \n",
+      "18    19.0      0.021960          0.993750    0.145674           0.953   \n",
+      "19    20.0      0.022683          0.992875    0.131017           0.962   \n",
+      "20    21.0      0.019511          0.993750    0.163879           0.953   \n",
+      "21    22.0      0.020723          0.993375    0.131990           0.963   \n",
+      "22    23.0      0.015733          0.995500    0.115951           0.966   \n",
+      "23    24.0      0.016863          0.994625    0.124647           0.964   \n",
+      "24    25.0      0.011906          0.996750    0.095556           0.975   \n",
+      "25    26.0      0.013877          0.995000    0.075135           0.973   \n",
+      "26    27.0      0.014348          0.995250    0.097998           0.973   \n",
+      "27    28.0      0.014791          0.995000    0.099138           0.970   \n",
+      "28    29.0      0.017454          0.995000    0.108553           0.967   \n",
+      "29    30.0      0.016810          0.994625    0.091436           0.972   \n",
+      "30    31.0      0.013777          0.995125    0.099723           0.973   \n",
+      "31    32.0      0.012082          0.995875    0.104254           0.971   \n",
+      "32    33.0      0.009674          0.997000    0.081663           0.976   \n",
+      "33    34.0      0.010131          0.997750    0.083600           0.975   \n",
+      "34    35.0      0.009486          0.996375    0.094651           0.969   \n",
+      "35    36.0      0.011784          0.996498    0.122534           0.964   \n",
+      "36    37.0      0.012858          0.995500    0.077891           0.973   \n",
+      "37    38.0      0.005532          0.998750    0.085279           0.978   \n",
+      "38    39.0      0.004174          0.999250    0.062567           0.980   \n",
+      "39    40.0      0.003725          0.999375    0.065963           0.977   \n",
+      "40    41.0      0.003737          0.999000    0.106305           0.975   \n",
+      "41    42.0      0.004466          0.998750    0.069676           0.983   \n",
+      "42    43.0      0.003353          0.999125    0.063304           0.984   \n",
+      "43    44.0      0.002823          0.999375    0.082333           0.979   \n",
+      "44    45.0      0.002750          0.999750    0.069678           0.983   \n",
+      "45    46.0      0.002190          0.999500    0.122289           0.971   \n",
+      "46    47.0      0.014585          0.994625    0.117514           0.973   \n",
+      "47    48.0      0.003940          0.999125    0.072986           0.982   \n",
+      "48    49.0      0.002838          0.999375    0.057956           0.986   \n",
+      "49    50.0      0.002429          0.999250    0.071051           0.980   \n",
+      "50    51.0      0.000884          1.000000    0.063533           0.984   \n",
+      "51    52.0      0.000639          1.000000    0.062502           0.986   \n",
+      "52    53.0      0.001002          0.999750    0.085928           0.977   \n",
+      "53    54.0      0.003981          0.998750    0.142024           0.964   \n",
+      "54    55.0      0.009453          0.997250    0.112729           0.969   \n",
+      "55    56.0      0.020304          0.994000    0.221027           0.961   \n",
+      "56    57.0      0.020196          0.992625    0.074609           0.980   \n",
+      "57    58.0      0.012778          0.996250    0.084888           0.980   \n",
+      "58    59.0      0.005098          0.998625    0.070511           0.983   \n",
+      "59    60.0      0.002206          0.999500    0.067209           0.985   \n",
+      "\n",
+      "   optimizer  \n",
+      "0       adam  \n",
+      "1       adam  \n",
+      "2       adam  \n",
+      "3       adam  \n",
+      "4       adam  \n",
+      "5       adam  \n",
+      "6       adam  \n",
+      "7       adam  \n",
+      "8       adam  \n",
+      "9       adam  \n",
+      "10      adam  \n",
+      "11      adam  \n",
+      "12      adam  \n",
+      "13      adam  \n",
+      "14      adam  \n",
+      "15      adam  \n",
+      "16      adam  \n",
+      "17      adam  \n",
+      "18      adam  \n",
+      "19      adam  \n",
+      "20      adam  \n",
+      "21      adam  \n",
+      "22      adam  \n",
+      "23      adam  \n",
+      "24      adam  \n",
+      "25      adam  \n",
+      "26      adam  \n",
+      "27      adam  \n",
+      "28      adam  \n",
+      "29      adam  \n",
+      "30      adam  \n",
+      "31      adam  \n",
+      "32      adam  \n",
+      "33      adam  \n",
+      "34      adam  \n",
+      "35      adam  \n",
+      "36      adam  \n",
+      "37      adam  \n",
+      "38      adam  \n",
+      "39      adam  \n",
+      "40      adam  \n",
+      "41      adam  \n",
+      "42      adam  \n",
+      "43      adam  \n",
+      "44      adam  \n",
+      "45      adam  \n",
+      "46      adam  \n",
+      "47      adam  \n",
+      "48      adam  \n",
+      "49      adam  \n",
+      "50      adam  \n",
+      "51      adam  \n",
+      "52      adam  \n",
+      "53      adam  \n",
+      "54      adam  \n",
+      "55      adam  \n",
+      "56      adam  \n",
+      "57      adam  \n",
+      "58      adam  \n",
+      "59      adam  \n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "C:\\Users\\nicol\\AppData\\Local\\Temp\\ipykernel_9880\\3072135520.py:12: FutureWarning: The behavior of DataFrame concatenation with empty or all-NA entries is deprecated. In a future version, this will no longer exclude empty or all-NA columns when determining the result dtypes. To retain the old behavior, exclude the relevant entries before the concat operation.\n",
+      "  collected_data = pd.concat([collected_data, df], axis=0)\n"
+     ]
+    },
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "<Figure size 800x800 with 1 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "import pandas as pd\n",
+    "from matplotlib import pyplot as plt\n",
+    "collected_data = pd.DataFrame(columns=[\"epochs\", \"train_losses\", \"train_accuracies\", \"val_losses\", \"val_accuracies\", \"optimizer\"])\n",
+    "dataset = \"mnist\"\n",
+    "# \"shampoo\", \"shampoo_diagonal\", \"shampoo_momentum\", \"shampoo_momentum_diagonal\", \"shampoo_heuristic\", \"shampoo_heuristic_diagonal\", \"adam\"\n",
+    "optimizers = [\"shampoo\", \"shampoo_momentum\", \"adam\"] \n",
+    "\n",
+    "for optimizer in optimizers:\n",
+    "    path = f\"metrics/metrics_{optimizer}_{dataset}.csv\"\n",
+    "    df = pd.read_csv(path, names=[\"epochs\", \"train_losses\", \"train_accuracies\", \"val_losses\", \"val_accuracies\"])\n",
+    "    df[\"optimizer\"] = optimizer\n",
+    "    collected_data = pd.concat([collected_data, df], axis=0)\n",
+    "\n",
+    "figure = plt.figure(figsize=(8,8))\n",
+    "for optimizer in optimizers:\n",
+    "    print(collected_data[collected_data[\"optimizer\"]==optimizer])\n",
+    "    optimzer_data = collected_data[collected_data[\"optimizer\"]==optimizer]\n",
+    "    plt.plot(optimzer_data[\"epochs\"], optimzer_data[\"val_losses\"], label = optimizer)\n",
+    "\n",
+    "plt.title(\"Comparison between multiple variants of the shampoo optimizer and adam\")\n",
+    "plt.xlabel(\"epochs\")\n",
+    "plt.ylabel(\"validation loss\")\n",
+    "plt.legend()\n",
+    "plt.show()\n",
+    "figure.savefig(f\"diagrams/diagram_metrics_{dataset}.png\", format=\"png\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "65b2c54b",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.9"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}
diff --git a/scripts/staging/shampoo_optimizer/diagrams/diagram_metrics_mnist.png b/scripts/staging/shampoo_optimizer/diagrams/diagram_metrics_mnist.png
new file mode 100644
index 0000000000000000000000000000000000000000..11cfa4b58e12cc1436aa5f15b791db8babc929af
GIT binary patch
literal 65314
zcmeFZWmuI_v^Bahu@eK8(5)z_2uO&OAPP!JH!7jD3P^{oh=M30ElPKHSd@Y&9RecK
zC5_}A3q0R<&U4Op@BM%850t&x`;E2cnsdxC#(E#g$(&ucntnBfLRlv%aax{2S;|Cy
zXjkB!#%n9C;=dD?;;NPk<_4D5mu~1&q%T=sGcmU`F}%!lQ~!pAp}85?QNE+c4l`Y~
zw7h1)&(3c8zdmr({Kge_?!?q+9Awos2{j7}Wy2-%LlYq!Zb+fbxl5itrD)^d*JSOW
zs5DhHG$>^!O22=VwCEtsa($XjiW}uNDK0DUN^*^BzjBGL;5Jj-MKMLin)?seB_1(8
z_2OQt%WDY**N2wf`<VrOPUH(8uqX);`K)(ePiU}~EnaUxaN?^^Ny?SL5r18-!W}fY
z43`acvrXg|%B?$;W#o<WjV|QnpPfrKlK+LoC`<6)yDPLJ<mKdA8d~yy5Hp2*>nGPT
zCi2qZ{{R2w|E~;K_;SCpFv;f!{$}sme@;fGCe&NIXVC2R){>GEdO<r&pN8@7%CnAB
zqmkdn+!zGa@irZQO-qzT%a@)KmLSDsgR-q#w}yNTIO#n-8G7FPvq5tAoee^NoH{=n
zR)k0<)tm5Jc6sDyFOf9Q8M<%6<CkXMGLUMKHD7eAa8^Jg%hJ$GTUKeN{HNA3rtn9x
z8tFY@erEFIs0<8ouU_pK>#J8N7n+VqN=lk(612HsU=Vgd+E=li{oO62kN4O49yi=<
zE+CX<QqL*vdt9>U>7J3lR_y925^Y89D&>KK-1g%pg5!0%p=pkt0oKQL3)i+-t*<WB
zbr%ei75e+BrHQgcG$H>ar8~J?s8UHqD>NbD*puBS<oEDf%9)wP<xLNqY_cd9oEodu
z85{n>KAr8S#LC8Y{{8z~L)S`6ORIjDJl`Mrp{4Mx<t`qRZ+KN5E*o^YWeS&2EF3VL
ztfFlzE)$-7E?&1jN#9q%HtCEzLu^Jy25WBIo4M(h`OjwQy{Wa@3RkaoNjBz<d_6hU
z?yX~qOPva{h~Qzc@8ejvhhI_b{@S5BrzWe~?w;s)WmVNzwQ;@gS29*PPBe9<j?h#~
zUv&6&@`-6<nzV|JV=Bhv&HUU%ew0prUg}_thI#d1{(Kw;vD95KhF>OJAu}^`^@a^9
zHZ~dg*>@<{qBewcOP@J&|IM2>TQ_h1H5jKDxpd|FgM0Vx{W+Ka=gI5WOv{%qw;l6;
z|NgzuY(?y!=_A+pFVemY5*+%p<(#Z+t#N#A$0mo2)T`e`t$$p9V0A)RSVtjD>cU7z
z8Qr#RVZVRtH961Eel95);`^$Vo?zK6Z`I$Rv}x0(p1Q=h+<l#I-G$^aadepQqgq*4
z$9k$6irf)usYcZmjLtK`bF<TItgPqO(I5Y|&s|`zuG`th@IbqBf_9^F0sK~5H*E^W
z*Iw{FZ9wRyHJVj6n&-!k_cyW~J}k9n`_Ytfx;UB>PBR$;IU_RrSy;&1t5&V*sf`bE
zn)`dj-QsJIu#Q%i<%QFy@AlNhM(59s#Wy5hk$se|kNBbI)O%=D9TjRf*3&(ZJ^1ot
z>%n$mg&%LkA6I2sT)Fq)L9|x(&8e=p?yjjPZuY$*=ioN`p?c}kC8O_e?lCyeT<>k{
z>~Bnu*3Px_3@Lfe^2H!oE#2|`?d4b*>gi?yR(TtQXXJSNq*XMp-zyqRF>O@-_+Z0C
z-8kk=Z9=@iA=PTETE)0FPOD%l<8s03bvurUBqk=hy16M1Hs_AbPPEvkTz8k9!!NPs
zDZ0A4+y<pP^bHIGhx2+W!`-p?UR(9nL|P9vldnA9b|j{#0Kp-QH>EKJMon3%d2>_7
z?rrCz)%dD2x_O&z#8kAbyr|tXBjxD>IumVqSJyqX7Z}d59kHCBonSk3Nb<sk3t4?h
zU(V$Wy58Uq@a53G-;iz||L4!O{Ra+^<y-On?O9p79}2R9`;lAroV>ZUk@wPvW%usi
zSMQu0Z#1{6R#NG9<N7LxNT%mCi^(3yimv-6HPPnKTN9f#`Rg&=o;|UMs+tCs(YeEe
zBje9yg)Tf;zh^6=2@|VPu2nMToY@&@PnDGR<4GN`o;a2fpEq?5@5KH3^-D2Y`ADi+
zllnx9^L)vmrdR*iSA%53@xJ=c#l;D66Hj=X4-PpXqkX>;<Vdbg)%0;QJC?=#Os8;l
z#D#sB)!q+l*sHK=wEHymZ5g~e{&ohbOpo`|?c5nTIcY<}cMvn3coSiJvVFuH$IkkB
zd(8_EkJrc_sm8T3?75@zi57o{1)2Bl6Puqb5{{{TUlXg*8yOzRS`!yDcqR1jpz}P5
zibDwP>({T-3)m#^nEmMbUO4NUXjpk*-7X#}4i1heh47~X8ExzJuUygSNA~38<UD@m
z>+asf*0>{Vk%6z1Ca!$1O~QC<)M|`||FPGVk$KACG<gY&@q?Y<4q+#U#OXf0!Swxq
zC-UcG&mYY^jb%i}-Iv!q?vO&*R(p5L{512rgj1bC&Ut(H?W<0>Bt{l?!PiXt3A1%O
zIZtD@&q+vp)3R<->X`4x4O@OWk{*HhaeerZEI>=lrWaqzc&(0e_44hSuOB|7<u<C4
za2J}o`e<j%aC4z+o^!C<+L@?JAMb~$B^&6Q2^&=%+*h)84}Vq8pATyceiuJG<GB+H
z?QAwu`zs8S-_%&IA8L}%l|b9V<0f_As-l#><c!s5bRcSuRfb7(ksmCBF!9IR3R|z)
zXv!j)8P+F-W)5WamX`W(m7$CoQGFgT*cG>B(Q_N=nPs+b<Fe?G;&Pm_wD@|WbBAO{
zrL6GSRhAMgxtgkIm2#;-f%=8(K?17Dd%);aXLR!NC)zFa^hSkGc3shS98(&ZY;*TJ
zb!UZ7db8aZBQDgIeI@>m<Mkc5P>zWv>!3#uA09)6?CARbw!rK>BBrcmZk%_d;yk+#
zTYOH11Sfmm<gY`6Bji4nW7T~!GWaZ794CCbE5rRTvh>+aX4{Ua>!p4vE<SeX(4jDy
z0RE1-=|Nu9L(Y9?pC0G&EBkT1Xiwn|-HOV}vh(b@dQ_h;?D;=R15VbGpBSy|n%zPU
zb#*2@BUry0ojGk!Pr5ak@#3D6)AsOPdndukb>(YHnjmskp+T}w>&FLv+jShilbf$X
z^{sPEv>Ccg?m5eP&@9kt%5bEVtNhiA7hB1D$}y^!Qtc;OA3UL2;GTWi;xazuV7+$S
z-(_~pvvhK(P1ZJOY@n&}_|<Rkj+_0c`+?d?77s7-7}u#&r#f;cnz<w;B@F-#3<}+L
zh9M>n@bJ`xgofJAl%sIT_vH*Y(ap}g(Q|*^p<9YrDC<veF{AoS59YUIIZyOAUKxu<
zGOI~E|MD=0rxX^T^tp4cK0=OJnS*)LQX~wjqr6H=O9c#r%raV6Y}virr10q;V}mkZ
z{ej#G?Jab4`d@{nO`iMtY4pq{>lUroO|CE9A}esgL<gf>y5C##vesPw{IsyeP{DS;
zA_n`?nO>cFPDFR^s@1DG#U895<1?#@xNzX`;nL&g%|FPMmilsk$@$yi7doCRP~M11
z^C1C@<@IVOm%%HX+NZc!!lLUNv#uKlQaN|6rp1$IdV#hbJA0jEWo3EuXNDiITDOiH
zwS=319UpGvmzmQ8*=4q!0V+MS3X?nmJ!57AKQhBq;<d?%vzjqiFKzM={yu<Sepyz%
z>ZZbTle)U4kBUAZ8V$27yDyUya2Zxmx9r?$h>+jyKJsGBR?*X2$EiL`dVZ=lzZ9s)
zI4*zA&SJVhUBmSA$jr&<zT`4QZy8s)fNr4koUO%o@$Gc>j>ihE2Up32S885+znsL)
zF0+jHrT!->*U}#^mf*B*w!M-$pC9@c@I>m`wFEK+w!a>+#MI^>T_yvVT%OR!RF%Cx
zb!TC9>da56>?_f<Zn9F&&a~+CMb@m0R$=c=Oazd_Jxut1=#1@~RU1t;0dA@p%W{s(
z9arC1Le?jOfA{(anO$!lOX=x_WCAmlp=PsNG@7RS{QP;5%jVBT_vucO^GHtF$z$@$
zcDRIPVX?$z^|0<M`2pC((7!rAtp!V|qhPzv@$H(|ZavcznxC<V)&jsJm%nPm23}xD
zUM_>uV?YNSSdC{q)spl|B<J%-I?H*H_&B$)$Q)<SpGnG`pPQC;aL57ReN=)P)6n0P
zT}x8Rr=p^TG{zF-P?2I%Uy&8~VLsXQN1<zubFc}m$jm`*?rI6v@edb#xs7X(k+6*Z
z)HxxM?2_!rpP!9SjYSeX<2km($c;nC>2K&WJ%F-OB#M}RGJUSOOc^aZ<_72I${RW?
z`O8S`ii(Mqy}z?Ew*Q`R$WC1;VA$?&!FMuIodnX(bKcBunwjd2FEG>iPN8Dm{jlqe
z4Q<{IKzM+e`MqV=xA%Jb`f82Bv!e?8N~HYx66;KDe?L1b@U?aB^>x;P>EVwXgj3Q&
z!@_(?fFnjK8+i0TZzYS$X}X`+y=^diP-x56t$H^Hei*uObo_jOr@$;;gCxR)OCMvt
zTnVsxIkq?RdfOSS1V+cvXV3I%;<Sw2I1KQS0|yTl&x{1kkMh;x=+cOz)TVAr-<L-;
zd;tM>g;yG;ekm<Ij*`zkJ2jS^mf5m<emp(j)_(-ZoY$(S>O{L^O}4c;!URE4lZFqf
zx!+=Rntb0d;H3FAP_VH<`8(y)j@n%+hUV8omouTpj@m4>1*&;M9qCIhf7uXjo<Cb7
z!HIv<imVQ6W+Z&-7q<e=s%tZE|Lk?fla6nP!pvWNHx7kx8Ez!wHrqZY{}$^O$ASiL
zOXruJw!4reo#tn)jT;+Mjh|56FaAZ7ba?!}xf@5hz=+$k-$T9^wzb>l?&3Cj@yOMc
zwbGV7^S9?_MvZ|Y(Z)1WX_W8O+4?I%+@7TvgC`vA_6%Nne}29XbDSP$F;xic)RA;|
z$7pPyzWhaHa8;1Q@S|sX43480+<O`MEh7@^O={XA1m;^jEeC$oQ>oNf0FxSxWQ_Z6
zX74<4BP6xVQAhh`y*@)x%Z)Eb7~<a+?!fG|52^oBR;S4OOecrbCVfCVb@{Tcu5`ZB
ztdx^e9!DV^wZ8x62r{u<N&s@xGrhd|IeX*Al4kSs@)d1ez<iv~o;^Eo@?>K}hm$G$
z487O$=enMro~2$Y+J>fnK#0%unw{qzjWy6`7=C{v?z6<hb1K>4@9^bX(wNapGTfTK
zwREMFiHS+#yqc}Ku6Rqn9M9VK!*5ekj!Vj0TA~qkp8MPB8FIa?(AH$WNO+D}vcO0j
zqZ~fal&#^NZPj}~l7KMF!kt&PHqI3>I&aPX+Gv(tHq{tVQC3zqCUN4->C>vtjDmKG
zk&%%-)zK=6t@55#2p9mifvxF^u8$vAy3(<Y0tvm$m=6zfs=yTyuw&z9nETuFd{vmV
z-{xL}IkZ;xngjU$qx4KF=}OSDnQ1+J^yqu>vwf*{V@AOHvYy6#s`1*nTNxSSQ8Lto
zF@IhkTfgzPI9Pn{?XHgY&YK<Akc_)@74@JRkTim%@Wy<+Sx`_AiLZESMmmL8MxsF4
zYMRT-d#b0IMD9QTQo^$)M(ynpjkK$u9&X}ym`tcX3WT{`+a@><1?aM2ujfEAnREB1
z=aEX$FXQs2jCyYXR?}_Y9-d~}sP2BE`?tHW4he;#k!xMwf=QP+=Fe6#^LK=9*RE)%
z*(uu}gQOpdRZoqX&OMHU)kY~rr&#qW8#kf}ARrtGMx$}Ha6Zr|at8BQ*LShPSC>EX
zQ@cC-%#T)@9xWe1o1bUYoRevFV*h?|09pO1pv*ND^Rt7_-q9QLa^kI<Ink)Nt9t?h
zOY#p*#uSf4BmfTeH|J`gkhL9}mVeo)(>kwfWE8c24}VB=u6?h&@SGf877gXu$W8*@
z(1CbBv)9y?;XKpE$XVF+m_09{L6UBI9wDJ(Todyhl}su7rm3+#Fap+-%C8a9SB+)N
zIL}WwGt5TNn*RvekCaWOFb<_E7e!ZjmW)MpnyE&rVP#<(cjxUjM?FF=$Dh!5%=Z#t
zJX_?>SQ{+%0QsFAeU=fBD1)gnT13(~*0g{7CKmvF$&fxcQZB55JjHF<C1<RGN@bQa
zRCr^8g`HjY!3IWUbYfe#Z|@wU9|Jt<Wq1Djm{e$Eja<92nAUKw%qP|Dv(Z^^Xq((G
zU%pJu45Zah1XjT+B*<RZ9sz+)$Mp#d`^=Jql0d?qApoh~8M$2<R@qH)fE>0axF|2G
zt=q?$(Ms7dw5Ud9zT5?c2HUGs3@hc$buP&Y4Bb8Rggz8sj6Uyulr&!4H-1Y*3lB6l
zAQ~?hr5RS90XhvRN<(iKJ^tZ{kkH%<-LWS1l&cr|#<%9NV}+;af7u2s5?42t0eJcI
zMJ=i0IMMn1`MK;gvnEz_fMkK*Dyd7<9YY3*x>6pP)Tq_Mh{`TrxPuh<PZn8gJkm|-
z_euh(D@N7^`T9m-GJ0=(FY>T&L30yZWNt|Q)F<@#$&(`^ZROss&DQ9dkEo~cbt~p+
zq?uemTx{OF`P>iZsQC%P;~-6_jj<H&C_{G-He)oy+<M?g<fAPtM(Di@2Hw75Ja*^K
z9khFIpKHf!*aoJHGp>}c4-zKf9vz<-1%PvkAo`;@e?+ye7tqloyN>NRDru`TnA5K3
zZX0>I{Otz<C>>|VYWX|;Evnv#uhFvnE>6#J`L5k~pITgA2ID?-pt+<SJED`9qv|Mu
ziyWI_?RHQa3QWLka$o)Sx^BBM;8E&$wzJCUynRybJQisWAbrY#i`UqucAH}f_1&}&
z4eN~q9P&jB7-2CZ7x?s4Me3TiB<Vj!*QnYt@hT~=<y;)U>q<rkIkL(gJHA&OwF(QV
z9mhzzPZglgeKDD*Rxuw#RZ~;*<+qA5C=ZMkv>&&r+CVo@4oo4i<o0tdS$*@i?x{H?
ztxYbI7o?=z=#O80O^}CXk(brbn+k*egQoVaEH-M60E~nKuxz&N>;{BVM5NPg-u$LM
z`HHuyqT1xHhtCP}0&NnCRjZ0<uRs$YiL9(i%KnP=yHy9A?Ce)O_v8zZ8SAZ;1=eD}
zo<IF$O7E-Sgr;N#66IUu0WpFYKunR<COGlKg78g9-DBm>bCqp&raXS=ryg%Mze1AL
zO;6+ZX%*XyEYhu_OLapz-SaCL+S+}Nzm7T)AhVxJ?Lzb&z3yztxcLXUEhL_9uqL(m
zi3dnHb<%pEs?A;4EcH6yZo&>d-F@PX{a~(&X+z2dMa38B@!uj8w@-%&9zNU`!r0UF
zc{(Syw2f)+UeOkZktZ(?tGxksCzMka;Y!{w-+mQ^F5JOv$}TZYp{d4Iz<srGV_Gba
zd2@`c@T?;FAiB&5v|ks1BvZmVj~eS>#YC-P&ymAPzmIlTR%cw-1#QxUTv7|PrO-Fj
z`gr2*@Nn&Ahvtmts0`sC$8j~1w1DH%T?O;}Wy1My_C=tb^tLhizM%*yOAxe_-NZmF
zxKzR6;(a71fB5)u3`gv3wjab|D)w{to}bR2uk5Q&wn8nAvY!}mee%Ss%%HMbMMo?6
zP(iNMuGznuk%mC5eQ0H?5uP#aU%c%@Zk$ISw*{kXTAx%f#UtrCKG+hkTl6#%uvNR?
zud=%e{9WtN`@iWePS2^v9j^%2)!N!R1;8fQArT~Ke={^5AY&@4{iOz}v|aCRE#>cx
zNgl=gDqz{BnDus)`NfVVH{!osC!HBI!<>aSIfit(?1aG>9km?mZ=49}<gee=vh#|6
zFxOW;u@<N4SX7V=#*;<ppTdy|4CiM4-r(EK57H(^7vDRo>*mX2YFEKjBzX70fdd^c
z97oe9Cnq;=PhtC#-c0xRV(Xl8W7&?kSiV{S{-F$&IkpyLq4yxZeOkI6(+fFd5JthY
zIY$jNrvlRZigomQ{W|$hxdaH8`EZ>@zn+rj4DyaHODH|cra7?BDZQ~j3muGA>%Hwp
zQ={E>eThY*@0M=3TH?7sRA^gt&sba%qd@MEfUB!(#h=!X+zyl1+B-UWT-oBj3D5tP
zm40<BShF%@u2Xpa0)XPS!>T`rOlKnR&nqdaHE$ubZ+!loI_lFz@$HNU85#5DUVrDe
zInA<@A{B)seRf>zhG5Z}E6WH(#k9GreVfVUB(oEpJ(#CK+PFtz56Qp+67_%s8#U7L
zKB^Ukq}U6VcG$+e9%b=fc|B%*7IRq8%zJ(+$m=%<|8K~oqbM{QjY-MLk8|0NqZz8}
zwU(?^piAH6CARKi&=W>sZNeP3-rtqtJU@4NIvueN3QpP65}o#W`9vj$rk4AfH%j(v
z54M29iOqk);d{a&OjK0V5(9SbhU2b_$@KzxKFxWK_Mytpt@w5qR0M;{HTw4Y7Pn#u
zfE3}WV$_la?z=nSkPR8v-_{O37M`C`Ko}-koj7tt+C1-X7{B#EOh^b-GA^cUxc$o*
zO8>T#bEr<A%geog)35ApRwkgNn>1j`ahit9nF3B)D_dHo0R9<B`}Eu_pi?<`@SwzV
z7TKfYN8C8hpq#&?T6I^5m-+G5g7gd>9UZ+~YY)_()VxP2&UU2ZoV4_qME4-N!Zv3)
z`iYh9!!gCPfJ%02@6G(Yvmph9#&d(l`|EZcc<?cpuyT#wK|wlUz4^nR*$Bu%gQ0?^
zvzBNxL+?vwO*LBQqd5xgq$45_STAqpGXl=3YE1BEv-o)?M`;ftW4h+%y*R5MbatTg
zt;W@>5dg|8x^Bb6!=J0YJsB8*fVKZFi8vl^C!2u`+J22sph2<%DQ2HvnSo$9>&Sd8
za3L9~6BZpd0?kf;+qa|I7V}$QV$b<=yQm!?S=&<}unjZXZrAiDEZ2U*s4n3WKXUZg
z@8_~F>kRM;Ji5EKZM)ayX~;J(o;|w|+<Ub9`U^k5Xe7Nv(az_Z^DUU3N?oHth$)=x
zQM<QpRqUJX10-U9xjY{iHim}tDq63av^y&|Z2KamQeb+z-GgQ*8Tp_$z`FT6mI;As
z?q1)#k7$Jb9qH7tcxK&GCHeNulM3W1lh7?Hahj31iL1pP%>45+Be@OfKr=@)Gvx3Q
zdBor-%2yYWe^3a_*O7MY{Be+kglOM)#_bz=fKjZ32v98yj0|7eKyn#XMLfH|eoqvL
zo=C8GQ|PG2P&w3F?%%z8w`-_~vGx^*E@>>ZEXueW@}2Vok)<I$^u}cm3^&X&W!{(7
za-JQV!Y936y@~*}q%}P;IAQZP9}ITN#rI3J?7Bng1?^Hm;=FNlTZ^gCW_0*-3z#|%
zz>F0Y2YecG(g=7_w(2-74aijquAl8uZLyX$(3sMDlDJ_IwClnkC<X~R3b^X|VO^hP
zbX44OK<XQ4bZ))kjpAp|q9mD4f!DNZd&1j8B#tJt1ptKmjzZHy&t>37&tnn+q(IV?
zE4m!#xe><rW^+0sOpqE<O~5m7xR#HP4=&sr7Zttvambq9ADPLwZ{Hrp!P&UEm7s~#
z@UIGu)XGjHCH~1yt~0pnTDP5C1Ufz0b26<;mlP^Lvn)xp*VW)e0H?MeqEb%-;C%-s
zA489BxAo#nDl_u}T^%MH6bIa}U86A+y*{aI19{VC{q@Ob(Qr)t`M7~iusMS@Fn!?l
z<gJwq-%!nBJv=s|W0P<f7=DO;X_pqVjS*Oz^}G3AlREr_o;!)5Wq)K;Z8njjnsaQS
zPxuqQuo?{XzFtdKFGmJ~^3I(*Cl7guOy84Tyha_RK3RyH&jQSntrjTQNJ*!%y2Jcl
z1O~<t2@qMT0r8OeRcV{Pt?Z8DS3SN8&pQbIdAGD7&oP_dswV{D)!i~bmrfKUwmXXT
z^MeF0kx4)~&YGL*_44uxC5sMNpP|<vgNS|JS~q=GtljY%wHW2Ux#HobeN~YPEYZ@)
z*`rV_jZtGrazw3DzwCHQDPwlL(dy67QzR1)=1v3_<$>1?4GHPW>W!-e!5WX0H>jSZ
z_Xzm)5Hb-4bpauPFpP}T%6>VWWk1>#3_OWpI)^~Zh9J@LH9#Zsk~oTfHF7rT9{|P9
z0nyd|$h3%M|0&Z>5CeWE&E?Jh2E)I@?PS3S8eOK=dS_U4eyw()S!!9d$NC_I5W=RR
zj{O=^S0-gcacqle@L^I=xo-BUKKJ)G0lg)SRNP&k9BvCbKrrmVmoHzqkcxS5vE0ne
z%wZ_^1|ROOVlZ{PF1<<B<_Hb1WI<SZQ<|wC7*~g3Nn<~I^)KjH)JLAO95~=hcz)fZ
z7P@Y>{i(H5UtS)OhB#ta8TP~n;N+3Q+LxA1_a-vKL(57^IFQ@iyQ?CfKYIF<2YD|H
zQMx~_p2NIZV1W$>iK&2Q(*g(bZOXQ`PdQ}j`c*;tPFF`B?rZq`b5GC9O<$3swj0~e
zOkU@LBvy&YC&W60M`OFuu1kS-J%`+_m_d0N5CRu>`&hbUx;H*wz4F$!arAKXUJ2_T
zH{znNk~KtlH>7$gKOWOk1b`8sW#4x*aQz*C49CETabVFAai1h(gP~sq1p|I>6vAH~
z-=i;yrKa^u@ol@U=27HzLjB#nf8UpgFAz7Pr7Hi>>y|>ZxQu0JfK@CI=5W}Qz8RIM
zR_fKU_k>!4&|^`u-`hs&t)W!6!cyGgmoW5^2lnp$aQf_7qG$LHwH5gx!H-ZUJg(^M
z_vz}o>V7D+iN~bw;?#J*F@#lpa_5Afm%gK6+4HL+?9GwLK`+PVmzMdmKAZ(a`>E@{
z*?T<+ssv$nD_@^lN(Kx7{$gBahwtS|CGU`!c<Xkb`@DjJO>R3lK71-D*eO{v`^)L@
zR!8k@9mdBp$!ae<FG!+2clH4>Sk6yMaz~*300*j|vfr@4Jdo5yd3kxF9!Sc_@StPK
zum##X@6S^Mk_~{07fiLcA(j#rG6%?)<1i0|6r~gtzL=fY@CE8>811XCJbn7~E5K~7
ztKXQCFMV9$&NDE510Dqfuq+4o`5Oou3_>clZp0LLtOFz^7Zg~~yOT#TvY&qhi3b@E
zMCzQMIy5uq;pb;e)Z<EEoh{q88T@K%BdfX;4g*03F&6v$BPEbG&F6-_^T|ArVc~L`
zu`zb7|De$Ea=byi!kF;HrTX4l22k~+o3gCR_E`V8B(2~SWoQF+-3aQ7E*X3>2~!h;
z<~*hiiobflqP|Fh@RTM(N?k&*52wB>SsAXg<*fF*zKjSLMb^79&xl=rd)B>%^qRjm
zE`1%Qnq?VD5*nKA`4{!8nLLzgC+;Hb42OT0@Z0=(jq~&l4p0t?toxtcM)d!@Nh|)V
zWog@9fhTAWX8{LXpFVwAW-wxH4R?~<U5+!D6O}tF*6F2>XgP^*bn!nkH!~s&<-HD$
z4%nLx^);lPN0u74=8f9F#G_IIIfGiel>fki!tz`=okT7e8X6k?$VeQ^T-Sb<a)bW*
zhHR9!1s4?G6AsP1Upq30MeM)%q=IpYE|hjlelI`2-a#`;xECr4>=AI!aMKen70p3j
zUKNzo=at=~)+#QQGMbwHph*rMkKaqYRm*nhx{VJsMZ;8qt9?~a<+7xZ$n`yFd^|%y
zQB7x_<=(PzpaZZ`prVoljv1V7+VeAI^QKMAlKj_xx`67~@l<K~<maWhnzn0vRLS7r
zV5fmq*X(Yr?69@9bsCT?I}N9oh|6V%lX9GayObUiN0*{KnKln*v6-fI6n&BZ!3q|^
zBQ*JQsa0Ry1%xIMpRrD(0YQxC39!ZWMFaF?L5US~ec1UmAR1+Xm7iZ7k{;nbxSOJ=
zN3^mOLCp}-;H$uhl=$h>T_DgFU=)B>1nL&S5GYkIe0&V|)P~D4Hr;GA&FW8U+=>c;
z@aqYFz~ISIXk;<kb%9uMhK7b<U8uqUrGjg6ZyaxZz(|nE>NRT=Zrn(MM%f*ev4$pp
zAF;4ie*jTvRmNT0W$rxf2Wlh&o+Q#hB?xt~t0kS*NH?qY+<%^E=($6MyX*jZV%kQA
zhmS+Wh}FtgCWJbHS)|j?4kRIr(oX0Wl&(;4v4&VSc4(4E!O;^Q+l~~Wr#@U)p5DH5
zC-p}AIY9_$zEC5HTx>iK{&5U4vmrPq^*o16Hx5Fw#Zn+JDFIZyBJ3;ych3(<SBPy8
z{Qceeby$MWht?A_tmsux>CreSfrZJ>UP_4wMq`y~+Q`F=g^C`Uv+y8Wd8>Ec>ATCk
zYzu>foafZQ+I|PHt44R`?nq|3BJ_MJ;cc7l<I+oDHG=%gHsk<#=Q+VNdxV`diOB{Z
zv%-D!qR!OZ+pc|35`e+?fYhtkD3ei8>^(yWFd#@}+fmJfC0;Zq5Q0v#V=BFAKTr_V
z6E5CDN`H;K7@-ucJcS;;`t8{bk`)k_34uYHVPYvE65bvG8#$0cNmIPYR#48b5NFjF
zJL?bUU4_oBZW~c=Pu$3iRdkfCJox68{LwE~<~AYud&RztUk^dKCEas%qVCzT!4{qE
zT9yka9C$=h9`cdkwkOf(S!&{S%(b*rjl7_-hmbZFa{!l$X9&8uC<v^|*#JGggP%T~
z7C&=F{rv;`#?M}Cf>Yz(`7;IkJbIdQvx(|`V)uH}9~l>5c|z{;OU*~r6RQjY>Yya)
zD_3k_=(`R{$)4m<Pz#6eKcK(ItV*H0R;OLPx&k(=P)Pj?Fda-wJjyD;*`6WD7VhJ~
z=c_@Tb5`X$I~RQVM0lwh)SL*ylzE~%>b<VnI#Oqn_B?RA$lqr)G!&+H;&+!Be%v=Z
zd$XziQgVG2{7J+Z2iiyxu@fWmOAqNg1Z}!;L&`#4qt<RbY_67g`Cd^w;kGa}igI!u
z4QZy)Gcyjcs)-+TX=@0xQhFYjwcsN1Zvz5UiB^s+c6wb%5+T#OdXqHH4R?ma>DZL7
z0z70vawQfQyzGLoX9O+__hnt7Th7<EmatUF)3v~`)`S}HMd7HOO0(#c0X~-DZL)k$
z+*X3X0yT$~aXD2u0yjY2EW~xlJ1OWqgt$@d>2Hp?JI{!L7f1_}@_B<0G6rHKJb<>t
zka+yqCO)lW%hn*3tsxj6Jv1Q?q7im>f8MVg0|8<D^WCj*fi!gRWN{yNo^v3692DG2
zWU%(sJd#qtfD+2q5~#W7>B)g_&|j|hgN;ff4HsM$ihyt49A~hmGP!d=M}4gz=y;;p
zUKWkkw@;6@j1}*5f5CTtyId2;ni|8qYIGDz-nFEiPI6B06<t7L<c8DG*?PJ5hr_B$
zkJtX!su%8Duro?3k4s1GubM0EQ@MnI2k_+5CBhizY)uoPT<o$efrfFtoV&IC)sgRI
za|6V_ykr@T1m!X9QxB?shyGM13*MsLx{QhDFeg>s^MzMXduJftVv^hFvg{b!Qcu~1
zQkr*)viaTKJ8L+ZZ1qI&=byXq4Y!5Mad<+Zx?EUF%)mPPD61}(-d;|LI7Hlu%wqQ^
zvv*NGGreBw0bS`7{4B&Rh)y}>M*C;ZJs5_;5`$zq``}$@QPBbvl>@;?z3yzjRPEw{
zXr+Wdg8~Mln;~=nTz!<tyA@1e$tp9hw~eAJ>qQ%*94-?;!k=gcAw{Mwd0I#$>7a|1
ze$CM=U2zo<q^6v3QglrdtF2euUGO94ZBC#m1}jS9hP02Q2?pJ>6|O8|&!{540Fq>+
z0|nG@YPLjz<xzZNf<W#?Wv%k+h@%On-9wSnazL;U_aWqT#c-KD;1E@I?%Y{3M|_3o
zT~vsz3Bke2&R*WYYdxSz9ADqRe@E9xnk<1l8%-b{+<G2heSmF;53Eq%i1!JUB@v<I
zu&C&^Y<atB-<gA4BcJTApaHjpKY`d26--QGkW->zO_C_w5nYB#L%ODdq9Qg<PI(-l
z3g}<Jj9~nTe=bJMfZbGg_!ozKDoBLyXBn#@G$Q1Wa2r;fUbF4+H|TNP0LQ2*=RsHE
zqd}jo0_X$>F{a^I2!}uzU=`LN&kPJX6O%9Tzvj>On*&zW5ZaVjLK<393>73RU>-RH
z%|^&^N_Aok3`_TQE!a<qlE&6_Wl~0D1ryu?@Md}u1aF$$*x%Eb9{-YEN4w`b>8gQD
zECHRp8w|QDLczz$L9VbGDdX)0QKQ|T&8xO^V_55Qmv;tkFZ^1Af<zfBWc4xGSslK{
zqv`px(z|%gsz^No$)?IU>J3sIMaT%ugD+}qvrXGNQnqV`fN@BMfv_D7D1N#z-F#y5
zXAcj|E@l2F6wt<10-2zDI9B$GLkzU<Q=#cTE)ZA}<-+qhMI06#r5E5Y0KFGU7|6<S
zS;Bc3^+Oy^0iIwvqM9J`#hV>DfDl!5Or)m;k|%_CEaJl$w6|SX@B@J<XI3KXqhF4M
z9^C7B{^)nBTCMPB`_A^nJI_Xwl*et}tVyWgf%qitE4Jm(sg((}1Xdsv*)triNH^lV
z(louP<9ahhXLZ!5YDgSbgbqi3BWZen3H`Qt($Hh1h}rbBgRo*q`5%rgE>I<OsXqE&
zUfP-v@~HGgr@z844#I#F^nL=^o1^g6g|}yX5F+*|wotUnP|MAe&6@Z}KtK6GZSn>8
z)%d%!)7$22U!TcyUtgnHpyw|Voam0c@}KGdsjn%Qdce#7s&~8g;Pc+46n_S=d8J@0
zrIeLl8jHzfacn4>>||ldEuAHG_n965l(tC=2w^D{CxK>*c-U^Z947zBmrsBuqnGRe
zhXxl96qCzTAJRM-aar{!?kh2A%BpU-St-ry1BZi#H;4~YR4ikN8mawAI^4jH+=V+v
zKsD`@>_A3NABX3mm~>s}g<4=2!`hUBoFFh|T`oBB3Xm;eM&{5rz@d@a{JH9^9P2^t
zeI){;-^6Otz?tllG{AW3=Y4hcp8zAl3+wCKB3mNC>!vrx7HNJT;SZ2`?%{D!+pc>*
z{3=+~T%M(v8o{rXAag9hS=gBDf^Tytq{+}JO~u4c31$qM&}@$KmgRaYLUNum3de!;
zlZnq6cEy_4Fv>=hsT10Us5D5lT>yvGr1S%&zN~YCrU&IE)d{4Nrg-{my%^-@2pK{T
z5lsjZUKNox388e>ogsBR9pemBT}?Z1x&7KivM8ZKgb)n`$*QL*JDqBXTEtLGn1|T_
za8<COxd+!idi1Ek=*&EN$6~mVy|b+c<4}={Td!ShD}1tg{d#5O`j}bJ2N$q-Q(@MT
ztbj&b4bb6kfq<*zKdZL2Kb?v$A+e6fy5);qK)c`|2<EZEUFH!WL~CT{=Vs9(0j;KO
zE;mlDfAQjl<!ALDNv!5Uts3v=23&;UHB`Z+6LPd@e0G}HiX&C0$x;WBLv!zuTGnlO
z&Xa&@+Ytk-p`5zwNa+gbW5+P_#GiXO%^5ro;pYgmUlx9fw%3J1p-M78tC)438qE4a
zNO~Q|F)zZVth;0P^VZ4~7^rI_MMVV~8cle5XZhirz`0d$wPUa)M<Z1>WLkuy8;<|+
zL$mxS9lv{-!O3o0Sr+kWt%kJW@!XtFX$X+gyiFkp!rZ#duY7H8o6CEJ$|I)?(i<PI
z^JCTQRQN9RF?|Uq1%^H)wCpKhE{s{VVC5)yynXwwpo5Y`+rx|rrU_2|yu%M?I_Tkg
z`8)RkI!Km;A;am>^v4e$;IvTA+@-LJM1cpj-bDWDi+uYF2hd!Bj8!C?8?OI4O-caN
zjiE#CjVXrD*E0&mfV@&0Y-~O(dUNF|GvS#(cZhQY{U&BZ`S0Jq>HvkHn$SnWJVLZc
zEE7pjf(+qzZf@I6Bx^)c`?Knyu&AuB$zS=;V`~Ee?W_(@;ck#*eK{Mc5+8^z)>!K%
z-|mGC3+NtHCbNFp)0|vp{3GMKUAvbGX|V`|hLD>~T+48KNLHY3af9}fln1>4Yx>Wj
zEwH746|(C3b_yLVQFo#8z9o<qDZTbr!DCX_V0jS5JWX^;AE6_29QB+)(8MAJte#nd
z>w6$hRUpGb320Pcu!jTa;PK;%B-IlgkR)^LM;KCXfh#SN$bkSOc1VR|@R0NZ|9Md1
zLXd~RP(wTlP1Y@jKo(N~rcp?8M$mCQLx{mdJ;x@IU{YjPX%vv};5)5w6#H0raiwZ2
zB&b}{D4J#W^TetppCOzzX`43Z|2`ruDta4GBjsAFXn#w-4swKN$eN=WB2aIANMS-m
zNRn{^W1}27_hDiQ4`nA5M#2V4`}1AE9oB$`BVRX``U$_M;NCt3m{BXxEnC*>d&B6X
zOeo^&(DU1gFz-}$!(Zw&EjJUzrDxxP180aleKc?KGJ3sp8-ylblLlwqZoWS+ifW=&
zWXZmqj`E;*Le8Rlu<MBo&#6gA2_qX_fRdk@<&G&g!77=B${RbIe5G6&VQmSbpHS0q
z-JroqzrQ@dsz%KYR=P0^1o1EV@?C$!0Q2DHE8W8{5lHB|U`v><gtsmX3#}Wr;wm8N
zYFI)!gZA<9sUl_j6OJwt25aTlF!s$#6A=d?Q;lp3gQQ#ES#CWp@s$<t%_}M@2wPB9
zrHKAX@$cB45}`NEBbv)?jN>%kGTB#$*w9eIy;05xO%7&Q3GtSnt<XP)y-Z{@Av(-7
z(b<%g1n$e5BiY(y#WI%r7bWG>_BFZ-0>X;CW&$X_bm~%43HO?1_qr$)|3Ko{azAAH
zc#cvnE%NpD^6h|0#U2v+zxCxFuBV7B_eyK_PPYV7g!3q}qL<?Eh(diQzU=J5ee8Ec
zEh!Wk1#l`}z7^Z}nvN}Fd&fb?4)DG3p~M}`|M9E%{RIf^G6*2g$F@{-fiZJ%!I9Ls
z|DkxfF<Ox&yLWD+h)Ab7PT#CI;R#DzF0&8N{^<gdHo2Kb!bNx6VH|St*FE^CRT82n
zz<wYTE;j`NZ!v$wN>VfD#ir7O$xy@)W!>zL0*<lYJ$Q|eptxw{65{g5=N7b=l%2pw
z_t)-x16#Bdst9q+VN1>%^ps?m6x!=F*mOc5l28BUS<1bREqEoLkqCuSxFg?j6WylW
z%zkCc9pgaSBx{3kCGrR?IX&nXYZ2gJgd$1z1usdZhIwu%Ca)XI^ABHhw`CMSjIXeB
z9!lg}z@2WC8uB@`jK`v}y=Yp8>^UHTPMnEET_bDp>&Za@G`pgt`5;LfPJ>7sM>|{C
zMMQ+M`DX;W6}|Q=k&P#tW^OdGory_ke*F2@XSVp!3}TaAC<qCc?x27iK>n`PwwDI+
zV$_nCdvmZk*3Uez-iQ&N7wqcM@CF4RMRXX5h!eCTMzwLVfMYT@v$asuYG5x9E-jV9
zb<TQ*tlDzmD%RN)I$ldCXey%}OKBc4t1e6)rU@LE7Gy)CniwBAzh3zJAC@FViBMTs
zLiZj#(CA0O*6!O)bXd}_qG=*3q}|_k?^Lwg?mTF|yqlbU>B?We{02DVs@D3|tL4d*
z|9nTo-Xi+CW8c!nTR*+*?A3yYbWTtui6lh~1O!_^AoHl~R>X!Ia1k0|Z7ueLf=%#D
zlOCix*WMbUVJ}>*Rjzb_>B3a%r?#Xr1GejMdmL7or$N{XkSXW!i`D%2dZrx(ts%!Y
z87eRxbcyqxqe!o6(|}UZ@O`twF#7rS40=L_&dGnYQ4?&bHEAEj`zTQiVhKxjFJ!U3
zLPFZWBj1VhgqSjj!wLK;pYc2(pI&nPadd>H*u+9Q_5}^a0(NV8!gG-h4YHRxf$O6{
zpHov|6dLrt??~?(aTKmt<SRjf?@4QqO)jCttlgl4?rIi33sR`9u+^c52#EOl!1bp@
zpCp^L?r&g>Gi%D~#vL-$Z=@)d8@O@I4J_0AFS%**H@|gge?<YKPE7GniP(%9;WlR)
zN}g9hKrEs=66=d$c_T$rW_<ThQD>^}{k0M<k7=*#izE#vpFWJf-zDk{;`E1Df}2~?
zo_eQ&by(C!{7XD2+ZinP#!4{r<8>50*87^Im1?JGNur`e7%gME$f`}Dm8+Vg+259V
zA(d9--Q~NKjTfQgP$NsDA1u9ffU9!M=O8ApTrko{b=4kaiYOR^C1PlbKsFPR@ZR0K
z#b%w;5+XPBg*dsmtPnhKrmDT?&jFR}y7uO}_ixBI?vxoQU>yQn@buRuer#!?kvWO&
zOz4!sy&D}~ve71H`4S54T2yzV`lS6I6I}Mf&{6E5$gyOj9HxK<J0O!c^HYL!zp=zH
ztEw#$6>_*_kp&8aY*?-#qypK={PXmRyJ`uzMY#f6k!x>O++CnDqIS^Gl9jyh5tqdj
zh<(lf-b&bbiPC6ARoI#NclAr-O{|<u|AvUBugBZ@TcEC*ox|xD)P?QQih}nqE>B3<
zIFT<bwOM>&EbR!#vX@#s_JB)qGjcANCqWGVHtmfrJFF?5X9<Vy5+x3v<-?l$7c-Z7
zI6RLRI2$OO4Z7Wh{wam(lKJ4`i9O6UJUt@#vYZ!kx2ozXil?`^!ow!IyK*ZR4saR=
z__5O?Mu6q<CcN>E1s{8ry@GnOJQyWLa)G@egR(@`CIR$LKqucb8lpu|C`u1;4)MEz
zv?6=&l(^uH*FBrS0Zs5T)mM?1@6IlqF`>65NPMxF+{G8@7;gW2bqlZ5#Jm2=kX1|h
zxdr!I8bwoLMJ5Lu!$}$36ui9G<+sAZ8!hAwf=K`IMy<=myVUO0(v<8Y@3};6`9@QW
zX5}?@eIi#ygKt;QyrJ*&6zFJzEL4~d-1`f65LiQq7HqmC3IpLMPnGAkonG<#6F<}1
z_c)Y@(&{>j-aiGNcqnIq){y<56L=1q#D6pciLMt}R2e>N*M58KHyOo#BSZM33C1O+
zK@d#XIHyuw7V+Yb$8CC#A6vNp^T8pH)1SN<0|xgk0YCU65gNdPYxi)he;|f@AA3MH
zFou|E`?w0P66znk5kaIxG%w?AsY1SqBvPQOxpaXl#^_|R3kK#L3*diASsD14=&E#c
z9b>;HGnJM+$5i;;<=ySEp@Pt;I8R<(9<qQT{>0Gf_T-5Y`Yste8Rr`t-hZ~(aj*X4
zevWknBSa}<i}KG6NXVWaTjh||i^f4;12jf{(j_~TbH^Ku)sqdRA5D|}Q2}qePBcxs
zV!tghiILtYHiq{-+mBus3D2gE)glX*_?q>9YEXMqM?iEdL3pT<MJt$lb0CR!jnL5G
z2KiUih47F|UBbB(rWq|rUWRC!7#b(w_zMXMx!jLykHUl)!}(Cd6JB{p5{={n#oR_r
zfuuwAPlnpRe_Z#!c5M%v%oJ1;v`+k$sh5hNF0M{>8Xubf+&&Q}A$kE6pNY%;=yTo@
z4hHRDO*!<<FyK!eK&}h8Vpy9M_V3?s>(2%KgiSf-;E=y3F|0$Cb!XJE*zobjN~RMq
zcKWXmnpldFy~WJL#pH>dYJ9}fK$sU0C~vW8>mXV~-9oo#R6-YHpGXyq6=Z`ltmrno
z8Mkj|kz{BjTR_P^I^?xbC?-*a(L?P#n5JFaP2EeCh*Y!n<SK1==@k^J9-KN;$A%MD
zDbnn6S<5J-xpD&ofxoHdEkafUKh(%ZW!GX@ln-U+I;?sF-JA_qKo|-11Yy8dD)<B-
znvK$rW%1YIm;bu9ti&w&i=qGL(sC*hzC^qdcIvw2pQQ1|hKYj0LRM@oy7Dn`Md{=j
zzwte_KYGi*PcpJA*lA}8<<2N@KQjFFymuvPh0M&5ZEV=8V)~8#KjUahzFMe1-juYB
zmPFlZlobnv;%AG4oYaajUam_x&udpy_zYUMrN5AYE*)7&h?|fQI~i_!J@1s$JGPi3
zkC3<4izKbSBC?Q&Md+CEjEsAazeSQnEusX<`P#r<%f+0t2014#gjsbl%PApWX`XIn
zRz+r6$Y8Id2o|2$7~?PSV6W(6_!=R{WjquWOL%pEJFZoURP=jZDI1x^67W(^9LLzb
zkih>y{-tcc@PVI6>LX+Wmnc!x({t)8|CJ7Mk89W!mwq^&gYS{bNfY;H<Kh2ZB)-z$
z_R=}{4~Iw5l8>)B`f54b#3#|a^?&Pc(1#0WFDI$q6;~jBH*1N9zeD)m+nO{FpIx}=
zl=h$gnECkI{*o}cUh^Iia*1^0Hf{bqdLSYbcp{OW--?M`z!F@*QNH3h4<<jV3N!!W
zkh5-b5ox`*=6lo5jWtWi@X;|*Zha7wjaj~V=PzGZ@5L&(goak2X34*z@QAXkW@Aj>
zk1dZkc`lAPzLL7_;n@{;7Yd{z&617&#V!j;-i7rHWyw9dDwnN+%NJx~iqb8Vc-j*j
z%f4tW72ULeF>v&)3^y**zV!N+>M4W!DDNMwl?eP8MZ>(9`$Z_Xw;%~Fs0ySe-p2ea
z&VKb?$}Kt<J-V;|aYks*?t3j43cH9B{;0InbFm(gVjiQGaAS4tzXOr;oGR&ER2Imc
zpL!&gz&XOD?>2NLoebpCO)l>&=$mivwR~{>V7X6cxfu!56*%Zw%CE035>>bGsdv+~
zA{Xg`uXLWEGufaVA*i@{VZ8qc0ifuKP!H|>eR>ZG`k&Ivm=avSHPJk@XVqSaf<HV<
zHmZsxBn|K8EcAc1>hZ$x+3dt{C~tJ(d-9_5*kah~;{@y8d#<iovg#6tC=R8#Zpp@j
z$BU0<myKD=1b+PNB6oK^-ZtMyxn;tm5$u!Tue^?SVYS%oqM@Y*ymIA@{$|v*H5iHM
zT_ddsQ{<n^ch@gQBIhzD59&QS^)>%yk8a6EiDAt(|0{@TsU>e|0~TkN{Ea_xn}m$e
z;!@m$*R02vv3W24nY{G;aL#u*oBKlC<gopZI*)v#)9ClJJq@YR*k`hpjxH3mgE4lX
zGMGXmI^tK>F~oK8ol8{SPVG$$eMd#Hgb0A7P_(!84X&U30&gP8s>spjLD^LjlOKvP
zu}i~2Q_z6N8E9+}_~U{;=q)%UM|2j%XbACud@Q;Ng3@v9uI{}70Y}zMRv-B0)%|}r
z;N?>0A{8X0MJ#8;XBnq;lAH$n&+Le2$D{?80xkGei9SiJ%qIsj#E2>aq#)36@7_K9
zUNZH_-F>2<Eb_SZ3Tn38CjP}~(D*-!8&03wo^TNUhY)7OL!{dRkpkY2*I;#f2}=*5
zGPij?>Z0^Vjfj6UVWT0k{ylm631q3wTeb+M!EZ*K*r5#G*x&+ciIs)rEJ#p0D45mp
zI>N^DxNS=a9}#E^umdN0Ab;KoJdGTb2o*Uwxx|t7-@gyCuoPwazkJzoPeX-cQ!q;n
z&GjoHHv%$0TX(S4F?ej)LL!ewNbRK<L{+j?3>Nt#+BvEyjtbb{T!q1rtbmMRnhLo7
z7CksND_9WE2Bf}Ai6hb||8UcdD^19<9?p2Zg4%M4F8F*Lzem@%tIk$PpcHN?+!ZmC
zqnUt@+sVBjE>0C3D6(jAcakU}ScK<>HjpQmK(jxR>WC970yNB`#oIp+bnMb?c?E^o
zAe39Bz7@8&PA>KI>Xmyac={Ait-i7ts*(%uay`msdip40!i4fn00p5>qd_?73%SEu
zvqMy=3fqdTCjVT*(B*%ch0zVZlA*~V?=9k8GElC!STU8}MP{b3udZVzw_gcCqAJx`
zg>26v&#d^*q$GLj0oz9os1QWygO6JU(248^E(<uRf@e8+QV$6VYGGh%u%R&$=5f=e
zEG6=Y8PK8vVcmjS_qMA)l=Iwsd!MDiAm73>m2y}wF7-fE^g`$+To|;xaI%J4KipHk
zG=<;v(}N^7>g(xsz0@%Kj(efQgG`{}cM&HiddmHh2o3Ut5ExM2l06lm#fSnAr^*W>
zEbmq;5o%_SNYX1;QUIIv@dPMm>m^6wYl4KJgq=p^4|I=A&w8B6$r6p-Y;5>lE0?=Z
z_+!dY@tNGJiTweMe>8)}F+M*n+sn(1xy40lTwZ4++(rv_4@B5v7Rp8zl0+d$T>r2)
ziH)*YTbf}k0)Z>>jNaq6+G2ip;YEd#FG;?**r<?a`FL1frd1C7SloJ?g_CKq5qXTG
zb)Q~w`iB0g#j2o$nnXQxc=D4NGh6woh2|t`8B(F-&Fr$(qDdpW7t8;-jV|vtYhPt;
zF5b9^7aWnJPjB^IwJ5Adp)?)7SNk8852fqn2eE{IC>LM-1sF2?pAIE&L%VG;GcHGF
z2S8bP6c(w+7q6G!_ELzT6=V}WFVi-<SYGbn5c2YyvZ(r<SEi%SW2_@C;iHEHmaz%&
zY4X%l7kRv#2jslsT{r!-THm*NMl3}Ua+x6CwV{0(f-=mxK=<W9$$m+23zcKbHXJMG
zDP`F<0&Q+nKB>ETETm|gRk$HkZ->R)PmjA*O}^N^O>YrOxp)KnB$)@k5Zc2MI`B*}
zhc;j_kplyH#5$dy>gGOVl}<e*P&an1m{oqMD1|g@=D=Tit%-3h^+i<4O~C_eA6{8U
zyNHmsE%6Z9#?Uzx5YUsQ3TBd}#bOUS(Z%Yw9)Ilk>Y}&DLTl~#VNrM5g<4M;+>66)
zx*pfEe5c1lx!RBK1=q!U{p(1z0kMi*t6c6b_tyN+XIEZV%pnN~7=S{mx+=g*1ef)5
z|1zdqF;~|~oZ6K2oAoryLSu;%iGvTtwG5~w+?EyCjYwco77d4TkNWcO;LBFl({0v!
z{u_h#zwe$-ZRKfx4m$IjfzKC5au1CJQ?ikW%rd4+%q%ra7O#Eb@B>yNH!7&%#$zKN
zjBjmSF9PnSiy6l^+Oc@U#S$k5L`Sz6-^zIWZBJpa=9M^MI{*Y1`uiB=<qss6tF1Ra
z5aY3h=HbbudnMIu*M1~DYdfYUYp?OG(|7$3if%|+dFb##sm%&Q_|v<Sz<UAN>qD59
zZr&y@x74%qT7THag*!s;-{l&!99Tt<ZW&X}<-1Y^B+;^?)3y``oM1n`?EKz^fgz)A
zDQ+b4;qlt{N!#{?ELL1FnOzuetcd0R?Q_fitvx@ST?;EVVXt`Cq7?){=hgrfP2u9=
z#8O}Qz?=W__pRJa1C&L~VY8G*;x(K6Qu&5|@c7RPjFh#jl}OB5A~jGbf0p2l&N!Ou
zza?zz5x+5eEB`EjPZwSQ_^fYNC3$KQey+7HcNfWpLByoxP5WTc(JE1zhtVrSlqHTx
z%A-Zm$s$F-_a0>#gB8;zaZaYh;5f&~6Om5c^kj;*EwD=RDYSw|OH&6`zjCxpEL5iT
z3kdj-=3O(l3nuULVrCAv^WT`Q7Q=U{Hqz{IAy1Eh`$3s7-B!I}AiiHVjFYEN0gjOU
z(Gd{`NiT*jb?mCH+jq2QO8HKXa4<$-Gjl_d{&}Sqx6f=T+a8{zT&F)&erfBT&9O{=
z#=F!u;tYz)@BZ&xF0Vh0!-rI}bb|n~>l*_U5x_&J2YACPj7C0fk?q1GjfkIxsFdR3
z;*r=S_4(`9@#?auSE85)W_ZH_20g&HnY=O<UZ^pwY^X8V>}L!mmK=`khKFD85R4>{
z)$JbjPmgxL1!t-WmO%+5xD~P4V?Q?8)Js(vM}zx+Vl+NINRL?(L+_!xdR@|&jzBNo
zTBVhnx6to<M1%~AqW1r5q$KN!=j#wky1dOFyaG|deZF?9+%pnc1i_jLL-J5hkf2tL
zTd$mH{T3_+JKZ@csnNPX2-eVUyzDel&Bz`UfB)U!9Tv8mLglAQLMl%FdSTRmz)}1^
zAtUABMg>D09~qD;;gh*uztjHfUGasb;zj2{BOy*Uy#O&1v5^RjfT<9dq!u1=_ZLsx
zi2|Q*pZ=V#6?NmOq$m0v)*b)IvYh=)4=z^!57Kp28;E3Viw;Iz?`52yd_cAbMJj}!
z^R&{m{dxkDC!U$1E}RX0rL-(6LlnzSUiFaitp%+|MF0N*?WEk7Bc@N7Y#BPG&Yvd_
zDL5y11<)7*fq>J@8@Y84srOcHuX6c~pRIZli?uIl$xYfCSgwxaX-fpv5Gl`~^i?q2
zyu>`^c^s}I;<)>KO$wlofH6b}p4;#ix~V4FmjNk;N*>^s1;9)8k4HgwF#_qtVAsWr
zt;-_hsdTkvhyC_)Gi?f9<m{?x!P!mzX(gub12X;*u<1tPQ0f`if|X9^L7k}YXW4uj
z0ub5onSuvQk;R2MCfiq!W4B{;8LS~*&z=SEGS9hy2gX!lNkrdUvz=8y;74bAyk^W=
z3E$tA>c+QzZ`~SPbU$mHi+BMkUjH5am_N{M*FQgh_*HIi^PViH=50rsEk)dQ>MwjO
z7ZPKeh^rcV9H&xQ7p8tp@7CZHaL_g1p~lWsdMcB#h}>z3jIEj;p(R3Lf8_69Bo0i9
zh(6lPP`N!4iF-!@dn#^Joc^$uoca`H$#&~K5;YvrD?SC#5Cma9>=o!e;NFaHy}ln&
zLsN`h+cQn~eBXXkXBxOi+r$?9VEMg6d-+5+FShqLqNFLq*d%2Pva_ZgPTff2Hidl$
z?xx<N!0L_nND-hFxiRctW41sZ{9IRkfI=P;1Rd?L<P@H|GYUxG4d>Z6*rctXq)uj9
zbBhGid%QuUv4s8FztEt6j^OEa>BCZ7Eh2;`ypTtX5fiQF^XG_ZUg3$j6$cejoqAl)
zH;E;LY`NF9j0txb;b!!l2K^F}WHwcj&`{tj_e&P;II;Wb^F6Lf_|7koKLIKSmwwoI
z7+<hCm0IGWN%qXc5>Ctq#Fq+Jre_GW#b|7FavIQDCl|5elh)lpCHxmav*5cI(O2}N
z{D%GP0$knTV?$X2ZA2kids6W@9hLMSHy#Jst1R8Bz4AA8=f?XN$>?%o<j^y|fq^Q!
zZXA!1`?4W7^42-kl~fk@MF60Z8c9Rk*+gaU9Y%1GJdbU0R}oxJeQN6nF{HLw0l$!G
zsMB>nwD&{mj%Knm6eg3dM=XJINY*8{ZNxB~$Bb!}hv+vh0y8%wu4UxGe9DP`p&7lw
zW*K=x*by@r+;fxZ7TmYau6xLKbYbu*=FvPmJz%ULdXXpD?iC52hi3@(BI@hw|4laL
zzJop}C{{uvkXpXi(%?NXgA1>s-&zTzI?Ipic)N&TPQ-J}va07O*kjfcC+hP9qE~&m
zOs%Kx6zAijN}kxi_Wil?^D2Q{vdrTfTP8MPI;yoX9i;RTZ}WNvfe5k*8)DisDp9@2
zqo~NZ*7eKVxKR&%xUT8>qW9Y8*5jTbX!+{(o}b=E+Il(_Q%)x6IxylIB=(aIQOw*Z
z;t!w7I*GaK#tJtB#V%ZWzZ!t&z~M6})aS}1m6{%A&GizWYE-Pz78i*Mtb)-7<9I`~
zVxA%BhI)xZA5WQ?{gF)n=Jk3imA$Ot_j;slp5w|m#6t7tvO1baEP*)kh>ea|?Z~ct
z*y*m4t!>!wM;vy}*6{`hRo8q9d>)qY^B32pML^CP-JmCer@<Z7!qaYuA!Y0CzMh+S
znhkj>SQ9pe6Z7!QkB{`Pu`FAe6F-+&+r=#Zpt!&T#u$0gWWwkg`%{c~kho!1pM|%W
zM9k&b<u^(}#6-ek0J;bZ5kmNP<o(B{Evd%dZ{EJ0{q~l=IDn>vKaAFxd-NBe|3Z?Y
zQ!-t$&=5jdt_44mVqAL^K9logCmy+`u-37cx9K<58g&g&LMGLj7HOpm)nO}s<Eeq3
z*p3e?gjY}yyX5LstG+{hAWomZSm&pMnN=?>r}>BG!!X#`$<>1m3e_O3iT__x9+@4_
zF6{0MtY%WCct$<kY$UP+hF^-y=O`YUhvexb5Q@77&8nnOI6r(m7f*AEmb53VJt^x&
z;xc}@a;DFc3Xs4;gK8sjI`$V0?IP2v7H&oPu)0&IBXNCk$>VQNI8ZS12AIjTXIAo4
zu!}A&KYODppu}<Utpc3qK&N(=DkwX>+$$E5j}a94tLSHF)4CK5iQDDAdQX8X?};ax
zFmxXia1=R?WlVg`RL#}vy{QEaocnFpzsDICIK9{Mv?4zvnbvHMUMneoGw-9WwUnyc
zKOqnLp8#Fet<#OGmw0HFQ|?xO!@1;kE@S$m=t1*{r`yx@%72UiU+k6ltVUoN2)~1k
zpiJSF-HC&6TTUp%8@;-nD)PoZZmop!v$eDf(%eGQ;j|srBTrKx{1~1UQBJ>ADHOoI
z`*5+#Z`&1ii^Y)SGd)l+#D{K)hqwOCX5-d4wN4_VErYs{zusj>Nw1T@Bf8+nYQ^9?
zkY?&OB_xGebusfk#9aQk_HiBAaWd2N#5W#<%;6R~m*2nb*DuJgs6Q7>?_Y$-)r(EI
zHE==45|hB)yEc8W9A^dDDSzCmCn7q$`u@dsj)f>}RzMVzCxW6Zdxl__PJu4t@@{QK
zxuu&uy$?u!-b=&;m+eZpbeL}E21G9&gdctT+C)}gn(G_h^9R~%t`8SGHO|yBz5&xy
zt&RkH$Yf%msUg=sJwN*b7kK143$2|VeFEWb95rUvl)aLM%T4i7mn0HvP;!W4q^?et
zIM?vZ-V4|u1y_~4$pukG(NilvX-5?2Tw}IN5LErn{M~cMeG;X^BnvRu{G=gD7ulRb
zHmFJxMh}k}ppv@}nx6>|vR4(`in9+V(}iiaA5^TjU4QR^{|+O%|BJ7;fU2rpyM{M{
zln4k&sFVWI2nYyL(jncgA|fCmA)Qj95}OjG1?iG*0RaU8=>`#y?(Y9udd?r;_rKpW
z#xuq<&N%G7_FDIS-Se8)oO6+qf1Gx0jl)Br!mKv@ez1ej!-3h4xZ|u9<>l2t+l%OJ
ziG5rYVF`U;_~_yuvnwd6Xn4_6<&frYxo0kia=!#y5TWpjRuB;%Xrzu|ueaI6HkvZX
z<IxYJ-~IiLzxNo<aKt;*T-F7GWKdshN`_e$`U!;#HRz!glJH(5W)T!~0rH`6#{7ir
zw!i5Oi1UB};$jQ|EB1g;esyl5|JKxV^owj|yH9C7ymWt82K-$l#sUaD5pOI)cEa|3
zhfInfkx>GC-pEMwc)a#KEK2oJV|#4>zu%Ub34axVBvq(BTfK1J%J=IiJg9hR38A_8
z&*#FP)dSKAXiGZC5rT!1P+|??Nnvi6QTT8Il%azj{NwjSgz_-M5kd>l0HX5-p{W_f
zGicAgC?0+W5S}p}<Fa}ox!fBLNs_;%;+F`NiqFst01D_jyebO343ztj0?g}|eBf3j
zA1g1R(!g;wFnRaIvMN{<p<oZsg0u*R!#_T9ux7#HAE`JO-V^vBeKzf&!~p&%^EA7I
zT^DYX*Q*PJpFVwBgc<SP-y3};oH>#+YnxtoIP@99+>}@$|Mv+v;~%|=0KH1t37o@_
zv+?h-(Cexx>tHiv$e~CslcYcRz((493xP6Gtrj9!ifE{S`11J@TcFj4_~0KSb9Q%1
zKtT<;S4}44IdO1fE_F_$Lx_;l6ms!(|I~65assMrqPwaQ(qM6+q|4_k4CVOwwOc5m
z|1gL(H8qWnib_Lpt}W>60y<9sZ7>T8>Vd)tnxqdPP1my=MdUb;2&E3;g5Ut{2f^>7
z6_Eb^-~2oNPDcU0rTF!0@Blj>Dd}GLdsaUmF&Bq)9}am$>JNIGmC`9~L;{6016Bj<
zNc1nn8mhwl77^D%G7gzlJ(y1r+0kP6#~>sg2H@yY>TAv(Xd@A(T7Lz6<_~uC(6TL`
zMxY@0&(g*!bv~jZeiiqxNN0Fre_a7%-mYa{>;LS~__~WE>mk7@NS}>tA<#p!{^NGt
zOD=$+4|1(`NNH);N+C(--o;=gnbIhuTwJ3^|0XIwERrx-mXWj(&;S<z&&(<yFlai7
zG%SC%X8!yEFRRvjpqEQKY|3+>X9v%nYg}gp$jWMJgdrzk)b!{-v^|){j3CO=4Q3o#
zU_~M6ih#xk`pYGI{SgbEJZoqG$e0A}$*kI$pX0{d{YPW@Y-R$6Lv&rxIjg}O9=2H?
zu*tHBGzmnwDPY%+Su*@z=Je3}7Tlk!05kzwM@+Dq436+W_=2j1xHja(L0rCng^_^V
z6^+t8LE=Ee#{qT?&cT&T_47`2+-=O<|3TK8QV5g-3n3!*`s;E8=h9C=wf(tzydbG4
z`nq#3y9eUEwn5Zn0R{{(?v$5-2D*P>64c|Gu#u682x5su1k^7Q5iJ_%bq8ikN=kZf
z;^Kux_v}$T7FVG4lSWv>e?J)RvwrUigy+rgKOZD38Z>fFm;gDLeL@b%pYTD33Tl0k
zvNaI4%FEwOQr`u%f?Pw%)9&k_lZY>|{{PVbMDUcs(%9N%@ex3uOk7(hm&kv9UX*Mp
zo;c#yx!;PIQV<pJU-p2?w9P3W#zc^@()`h9%t#eHo_ipq0L2I5+eETFpnVmY-C2h4
zI3!#R1WE1v^}P}e_jS3;2+U^xu~5!Pe}Q;W{mx%tkcUUujmW(~6KDY)YZbWH+db<*
z9lQV(e|Nl~LkS4q$33<>AT|0Y%$XSE0d{UPtAp0Mb?2uf(E=a@9q3dLULV?A7LoB%
zZ5cPWs@2LlniEj8{k2PS8Rvp58R>ZzC+q7eRxxO|GBH>@ssD8l%wd)qn42JeY+rEn
zB(4KjEv!XqD0lY}i+~LTNZ2RY{tw;syZ?04zl&py-Q>ZNg9Mbdx*|duBnbciFgvIv
zm&@gl6ue30_m32reaJgYUGL8t858pfj%6S!4(n+iZ;&mFW^BuxO1t$yq><-W2kE_+
zaX3y5#}^)#50@TR^?x4CR^kCUT5_XuPM1}GpG5)&92`b*qlwvc*Jd@DL8P=6#VrF~
z$B~sz7?f(XI)oWWeukpz6Dj!b=m3k@<)d{2bkD$u_<4}G#@XL_yLn9}rQzl&4*fm)
z5Q0$6Ctjb_9n|r};m=f=GIlEB5KSl;CVqm_dmAwV0|rLsXGrW~o^h={m@ZS1B@R8#
z&q<>^S9~*ZhUhz&x&C$9HOfl_a)@EF9<ih#d18P4xYjTdM9km_u>pzKUx`F=8*-h*
z0Xhl?J!+H`e*ZO~*}}HP_^(DRY8ZYg%QIMFz=c>8tpIz`4;2yt0{H*|&D*13Qt*%L
zXDt}PqJs-Ya5(tIt*qNY69oT_1lEY{<_5B6V!RUlPR#!4H-H>**MtGdLi`WV)gvYj
z2<Fq{H3sphGU)x!4!XRL$EvSn!WX0%2C%9OyQUBRXCkf#P;DmOq4RG6LegJAiHe@l
zzi$WmkjVMwux6QDs4LGeFT>rYsm=Ls@?i`cjR8p>M;g}p>PwW%FId=R+T^uzIgEx6
znytjgF#e6EqxvnVeb_-j4~i>IIk{j+O0erGLzX8X0`cpQ(&IIonYYsZJ8#K_GyE|G
z?LZ!U5;W$@LLK=XPfY0>vsGRA{v-A{4=CWbqmbh9RZ(cA)B&6P?bM@mO5*}4yi@_5
z7n%K=lo#B|<D~;G!!T}i0&tGjP$WtV0Tet6>=Mq1E~QI7|AtP=y_C9Tgh$U#0qg=^
zJoC0b$a}x!qZt6htRmR{(J<2xx~&hB8RcnBOXciE>4B~1I7mM*9uMpctR!G(zG!D+
z&&c6n1dl(_dG<6n^v|{aqIlE4py&>xrMPO<zYm+8i5A8~{blr%g^|^y24dU+C@3XB
zZb4R;+wzCMxLfNb$UU?+Z`b@||C2occpw=TNHdB^dl9=XXo;Qzg$W9JaF}j7?^%A*
z=<|{L*Y|E_0pdk0*oX&mX|Mo!S%c{Dx+KytLh%~_AJFXDYX5z0?6ese1pcE#{(cEm
zEkv_~CcAbmH#e6Z@dP5iJmi=Gh9{6!6BBKua1o{!=$ObqAC=)YBn##6W5_|ooz|Wj
z@ZB)l>>qK4SWj=KoF{DGe5BYkhAUke`Tuo<Z68Pae2!<iyt@%|A>6yg-><p;j(h*6
z6hXEhl1Z^&?arqGtxk9cGgr<G0nWc(kp^L{;PyavIioyq2-;9C!QGYt`f46bZ7n>!
zSYGD)FJwdQLuomR@QVLZJi>6}df;?hS_5_-SljlnT0FX`BJ~yw!QxPso0BPpzv#lo
z|215$njfx&Bf*W3LbBs5a4s}YmHx0{TCjPS)CU$wGV)pWAZ<nbcZeSnv6=uWe^{-E
z(V(r0L-LVeMTL#Vry>XenBOU<+w+mTa|PO#Z-{9Qp`nOGw!^)SH)IjF10n=R-p~Q=
z)$z&koJ#yBjUlCG^Wyq1ipEz7<m$fb$QTE5xno1cv&@7_fJVY#5x@tR*AHO*w%hho
zWWEfx#C7LE%Y0a2+--fQB<I4{f7u1#1SmeDM0#-LJpJ=@Heh%}e4WT_0CCF#=Wx2J
z-Z5)@aXcpFgb~(_I@0_;X+%a%;0O8Jy&>X85FYTs_aHWN#A^m~=9BT#_wCsjEP|NV
z)ez?0=W`B!?>ZRxRb@l3b%O+98%`ud#s>ilTh8T{qG4CU(f{o`a-bu<59bTw=LccP
z@p$x?NBusJquygY<b%n8rD(?UHk%ynm*#fwv-!hPf%75p*q3Y|&J`+~7E=AY?*3XD
z_<G-CTJ=qj)3|wW9f3#K(g!~Hi^^gdnxV98Mcqxi42FKq?}_2LsTQA<oZSSM#e7h$
z*nTxzCI)DK?VSwT{AJjBVm^|Wi7*cMBA#%c_Z%DoO6aJ#Bp^DHndndr--lm5w)&}u
z?)zS+breQXfcYHfFRhXp7dl51%V0KZc2IdU6AwPm@?sa7!Rhmf+9P@)WZ|h6S7O2;
z4Kps{czsMA6#CI*_6d+{NPC?8q!><ybHti77`V!LbeKIYWr{&zmua6gBlM?bS;;j^
z@n0f9Ta*s9v=L<G5r2@R$MI$<8%@BF|I5vFwI+bZM{C7}hKksQ2tEX6I{GW2=(PQh
z&6wfueo_MM=J@+63H){kU6AJLfyyk&{pBLeip|imj-ME^9{y>0bmX^uFNE-I%vYpS
zvAkmXhVNnlqB+~Zv+&sjnN%E>!0L5*%ZT#IuP2BGTf8G`<(qaL?_|||YFt(jdYO@o
z*L+~e%CautQ757pUCYN5eDBZ~I}S@o>xam3+uKrcRXr>^=1BdGS0iR54DuG<$GXqF
zoAfU-%HQpIv0q{WyTkHw-Wmg@AMghE;;iv380FhW^@LFtmj)%!+eh7#o+H2js+NxN
zKB&Cpll&Q1T=j|j^-(bP?d7qQ6-ayBL)pJUL8Qw2w0Yzllm?bP`A2ArQ#eEOH9ehY
zChZk5=(N1nvbHJ5f8V!QVgALwy)6S9D_Roe$J3rQs77e2krxkhN#iijx2acIgqaO1
zI}KNa>e^LApdFgfW=G(qCJ6>^8@ni{vBCWtg8^dRyF=#X!~Mefl?;I(YgCPd1vDW8
zAE@{SZ%{vj?c2}9q}MCX6sfj?WzGL?NiYF!&0Dx@F_fhFPM_kI8WXjecgx>6_xs*L
zM)`cQ-3T$936l3Hy+`-o4FK`-BPoZ`RCqTM1_;T2K&O)`k;)K(aSx`%$v;R^n@Wdo
zYlq@Ra2&mJoFNK3Sw%fIA5osYV}JY;MwY*<p@uwqoC=cIOjvADdQwlruQxXB<XHS3
zQB5Ocpy#>Aox*KT8B~^Gv*!uNna^M)FR*(c!-|$d`PpwUOL6g#LTHXDQbqXQT=tQw
zJO^X_0r-AmpJYQ4)5elcdBi~A>-X=M&7vu(^GiR`VSaW_nEAFfwASfUh$un(lej<W
zBFD6*wm1$QVmpC}0V0M4_-_H(eDH21pjVNER>Z^(P7Q#=90L=SAHmv_4e%Oq5CKV0
zYNpt6B_xWNG~b#@et4IeMNCnsu-o+SB$H1TfZnwh`d4rzH({m;u4oIyM+mQ)K$1#x
zKt_ZBXkh159~~j>%X_Q*Zpd*d@i?%D0iG^0#z3slrN~<c{{L0rM~hFiD!;zFMKxS1
zl3Qve86iaMjQV`jGmK}G$75UsZn*U7vMi{ON}H?{VKvX5e}>|L96$)B(?;4o;A4>H
z8A3xA0r_(koqz*0&%mZq8rmamz+Brfg=t%GM2JMR;cysei06@4G=cgj8jVf`cb^92
zmuH9`%#hMfe9n%hB;}MJdEjvVinMuS$b~SC+F&>IH$McERA&5oOZZm%v0NHM;DeOS
z|5-}TlW>EVR|5`)Y@^W~>o0H?mw=i$%%P>F1@u`q>506cF9SUgjaWuC{BZdml>;TC
zt07o)B#2F<j||q={ikmr9^xFbZqs}S!~u~oK<*oe%fnXSgaH(164&`02<mJQA8CJ?
z63rK$i-+i7VD0mgki6-=Py2=vC2W|R)_<t-FXHgaRVH#tA>^zoZv|#}BUQMEdEI8n
zeF@AcS#fN-XqaN&h6oyWBySKFknz+OWQpAaJLEtY43}z#+Os1701ijf-Za-teSEE_
z&WXkv;WT{Sr`VxLE?s+Qee;4}8}2t$+W+2p%HI_EUpi0S(ec*kx%mGo!vx0z&f~Cf
z@<^eTsCI-Z`N^pQ{}t@OMm{>2mIRDQGOFSDwc&k3F@Sp%!gPx8zcK2ap9M;eYo`L4
zMuCg8yD{KGKd8vUJOTJ?0bS85*xS<s$8tzyVzH0{J7u8Lolmu}yw2l&Xs#k>6^{4R
zmHiyHUt$XfvpYTx9a@<zH4@_<Bi3VJMiE&W%#Bi!d`kH5D&Pc==_j{oiyT@e^ja45
z=ejGvEIbzh(>fw=jE;}bh3$n{24H?23$9>(WR`*O7^M&#iC{2jQ#BzP3vpU_k9}n|
zz1S4;bZJEOG3ckl<@mDu<<KQNL)uBZd8yj8x_M^qF_2va_C{@diWJ4q_}#^u#m56T
zHN6p=cK!cZseU5mvD>4_#R7>}VEEj*(*<FjxxnDbqg%pg3t-MX?lS!rVJ4BRI_T01
z)t#<q+}P8s`S4OdGnRq+K(bBvy7WNVNJ+45%J@R{&S>|f<BOi*(}ky0D@v0&OZ7Fz
z%F}&1lJ)Af-`q}D;Xup?fQN+!&71M1LX2RB|3*hCjFkg{71j=>VX@f*2<8Ci=_Q1q
zH=Z9~XcxJ>mY;ho>jke4YIDvqt2sdXL(FrIQpdp_OK#EutNNNx!WmZlVJ1Xp*`g7i
z$D{NBx=7Q_U_cPih701s%sbB5kG=In`+l(xwut4J(;wEo@jQImE;}%LF-Ty)NC0l1
zamW+#>k@{%523!BL46$?v+`XQOXT2N{yn1mm?;-jMwF%=BlVju*0pR$m!3H=VhYeq
z%+JwapzIEr_5D{95gj)xytQ%m4~?G_9fz!?E^LKj1y-`B46O`oL*%*JHI6G?V<tTo
z$2Q(AUY-@RlJY%D2{vd3>pM{NQK<4qd~^+cKFOPK%Gv-Sh!bkw9yg@+`MmtXQF@73
z9zXQd3ZY6d4Az2fZk%%ahg=10U2z%3B%M7@EBTR1{M6rslO^?nj>q}n23)w`^g0_9
z-S#;YD{VeAf^S0qoL9tKF}!=q?nybY@JIX)hUfYy4m$0_-O1d{aAvR+T$dkld>koO
zjg9(Z(O?m;(B;G-Abv@x2E^#I<Rp)E5|{@SGLL%g(*n(_UXba9|1=8r5_%&3<c!Av
zPV9s@yH8b&PLT|ydDHrX{z4$CNqbX6Kvfy2!+Vhn`H;(KVvRb9?cVPVKVH5+8}*Ry
znbbJt{LsN9Z>cjH{e9TwZ%mByiL9>byGB35kF>hM#1rrn@CDU^6OBMX$cKh^h*(g~
zwfQ^~Dbpg%Xc{zP?_V}CXKRi480E&YIFVdkI-Y>>5X^I^5Q=0)vXmx}ZK=|~I<K)o
zBi9~5GNRCMNnJVH@o`T7r-zP`0Mgk8<8q8zoey$yXb8=ydyYHums>j$Y|1{<qWeC*
zxGLY1xGe%)@Q&gc3U$Ybj6*z@St7_87KjAcy-r$-)02s_ylxJLQcoIKFY{NwZ{>@1
zJS1pxn~oJg;96@Hb${_al`zkuheejUIVlYI-54llZWJilKp%E5fPu{Tq7hnu*Xd+p
zJ5KhlGxfI^(^D=fv3MulE&i@ZhLhUfNs-UAHY<x{eZGiI8P|^SV#8-`iJO>y3djfO
zTKT4#Rr8ln%^aSzs<X^h^mmGo4y?(ZBw@sDEnPFa`?<eP*?q@yDCaA`wG~C5Zy_B$
zL_+~O-Xp64L!1i^1}^K@D}J+`K7Cdk{WhF}_UV)LWh0drx=%!~mgIIX)s`QK$TwSw
z;Uf+)AH`P~pcVtE7_lbwp!T;!WC=4Op7CdOFy<!&1U@VccD~p!5>X~u8ETlc*SpeT
zGN}o}!jXQ9jIApPI+k~x=?_CosN24^uWRkW9bJikM$Z0eN*dX65{ps+Z8nkDls`66
zo!1NJkFD}LE0b$;K7V!R`XUR{Z5hJppJjd4UJFY_%cgG5z13*p!4qcrgkT|sH}|6=
zQb3JI{8O3#g0{BTfCXQ(Xf_30q?}#2k&|WR;5imt@hb@&(u@_KrpHh77$5=ZpqBWg
z<DxAqub76lRk>DDi>LB%;$?U~cMjM=tF`iN*ELC%7cQ((xzCrZ1~t5*{^7NeH^r5}
z*)EX0#cS3UFHry8!a^3iO>dFEWOl{XhQ+(gKiDc8_%J0E7T?3wX52gp$Az!Aayya?
za#E{#=}wuYl3@8mOO)y4VP*7l-;am7DFn`RGFAy9K{bDy;$av>MFX70zY9v?6tZW%
zTWLS_dp2r&0FYj8^Vz3I=@mH#FjzdzNil2lLKGq=H+G2X;MK8M>})8!*Be7N=RrOl
zfDb?@01S-OcZ``aNv<)KF3%mh_hz7#AB*p%t!*s)P(SGO$h%vijCOp|7-uHR38)nC
z6Zn!w?EB6xFrP#HB7*gX#^;cc!^XM{xD=nQj56heJEb;6Q(Yy`@ejt5?-?*P@EoY0
z3_HpHaeTiYKg;a=Ju^r-W0t}}VcibuO@YEE-;A_l2)NdKr{(WeBZunwob;F|g7Aci
z`dF#w)sDDFoSLrZ7d|ZZbmI`{jAn&mK&mEA=YT_tCZkqoz$Cf<uVuhjWX{4hLN}Tk
zmC~V0hWW0>h~amSW5~_<cg~6q)Ls_?aO5LCljtV^3LU7JG)KA@siVQnf)8FOp{T@<
z9zneh<`G)mCq?AnMlh`y<r9{iIQNR6;2MNlt0ms(`T!l&uM+cqZ?f-Ra*xQ_H6a+1
zHCI_P({Gud>ISt!*Us2*)dP|6aWRF~y#o)9@Dr+|9rKo?&!fu<S+VM?(^sYVQ~{Z7
zCLDOh)xyxyn>7z*bRC*3aKZ!(ufPCJ>oK=P`MbmX=x#=fUUN(G(fj<vj(3g<Lm$&-
zZPw+9&R_BBXRV8dexuE^MwT8i7x|zxM*i+tq-|wQAF`YexO^q>G8IXdO!r%cgC|V6
zZ3Ol94iu>x;7G;2ZT8)#P@9}J1cep-h(pFjIAdRV&#`8-F5<I>N}uok115H73^&(-
zGv*LP2;tji<Zt{41um*g7{F_Wlu|yFrwXs)NUzFcFi$L*WlOr8YW~N}vNu8r&B~7%
z^TAzuyF*rhZ8C1{Es!j#9J&~C^3wD41Pu~MbD~a?m*t{}E6x!1P#09V*_}SeZu>fd
zZzS;|@Um&O>Ko?-Z~P&U#+GG;!fl;3S~CO<ld|niYJ}28x*vriZAmT*+!%&ZsB5*e
zY^ATPnZEO-WwF-tu?Qc-J8HG|g@F&Wwk(g3f))q_*3$DS&98snkboltX)G1w!3Cs<
zCw?WKMS3vSK8)?)Cm~;Mzbc$ld1K|D+YU?qe7PbYfd4aWEoZ%Q>;j28a@QChiz5bK
zRTiM-V2J;sx|Au{ddpm>Ai^qO`)uXZBzn8mB3|FY?k&T@4>y*co8@vU<uP?0$r5v>
z(#UFkbj4Vc5xIePQG)RiIb$B?-{0m6g?IH=KjeaUVpP{haJMX-983?<B7^u#_ogg&
zI<|V=mmA%U>5B47G+5n&x+_3E%#gt{y9@WVDc+Hso&35KBXpeVzi&vAk<4Tl-+5=;
zb}M(rSY0<a>bbbl)(QwNe)|5v9Pw~09z2@v^m0_qlo+Vz8unAZk8?bH$6_#SR>;Ku
z+2_WWjnQ#QSqT<*YsX;vs;puLsk>}pPU472Ba;l`F#I(-{4ZZ)Ev=G6la<XXIu%SB
zp||UeCTIgo-OjquX`MmN^1BZfvWI&LEA#2TMzD1p<wTX0|CqZ_hLfYUqK*E$=Ltb&
zC5H!OIER$(<&RXVTX6L~B+6yv!B98`T2!2jKaw^T*0e<w@QsqoqdR|mj}7dM8M*kP
zKgCc~F?&#3Ve)22wRYQGd6U<6L`|tkVaMH!cO77zyV>qD^TA_QaLZ#vppO8qx&$QN
zQ9Ll?)!&ZxlSsc|(9c%m%V2SmvLh7b^zCkf-OSRutaq>S^rB?j7b^AYcI9QF*D?97
zviM<<ySZr5T1D3Ba(-?4!%N7?%xB_eQ?=uq_fx-lR~AcF*IOevGz5;>bVhxRjYNhV
zVsZ|?r5o}JLv{C87E2-{x}DwYay)RZ9rUyc=nsdHU&0Gnxs&TsJ<^euGZFsmg~iI<
zo)A4VA<roS+^i=M_%TQ0bs3pZyW16=fCwYteC5Oui<GQ1Y?>aDM{VI45z1sQ&o)z9
zpXo-c6MM|Z5|uKI?YzA|*D@D9QFn0ABOO+Cdb)`L?}^edmVRA9j$ahS46K=Z&z@#}
zJ>b|s%4OKWS2Z?Ld0-t|+|-s7lQxq6pqO4rx$p&hM_J^$y#DSS&#WNp9k0F!F3;QA
zhqIF3X;S$vjw-U2pdMPB+Q05wxbpo@8c~a!Y*|8ytDI`EeoJ{2vdxnJD0V1CJBFY@
zJf((*+<d7V+!HdY@Ju1~NqKZy*<_b`LZJ(sYft-qwPx$pGkPz393MtJCQ|GCIk;{0
zCkdU#OWG-XP{jYiae*MKK~H@^ZJ`zr*`Ykq-{YGZ1fp^a7{S$q00l4KgMI8^F20+@
z3KD&z7CCdrSP?GCx!Ef6-JG9}HIqu^enc1DHLflurU;uSS_TbjPtUhxh>9{8RvlVr
zX{S0qc)7iyl%M>Gj4XAK9b_+pp&D2}P{hf(EH09-RT`F_{Ou9&kq2C9-XIhcm$59|
z`u*zYN>B1!TC(cS#9lSSf_L?qPTlWS>q$Aq(}lyMo&`VwrAYP2pVEPpjAwJNEa)zJ
zxAWZ{{G^g@o5)_u5-`u#9!2{pbI`q3*N_CIpql@eSwcZ!(7!KL*yv|Ol>V}{eLc^<
zR=dY_4}xun@wsKHL)UtCOIFOPKTjW}5#`6e(OA4KA+#kV-?NCB!x;T=UrX(6R|Kc)
z$j)9*R?jJ#_Zr8v0qZyNv>U^+j&xh%2l<#8mouI(wy&75f&u1ibmgS^7DjT_M_g*K
z*J4FKDIQiPXc8LMtO?ku(NJ?5onL0xIpXOEoc-o}o|XzH&HqLImxd>i+%i@~`Oyqc
z%JYV&SM3JA`B7E9*{*L@d`3cZE&s{$4hFSg<8f=1(fn;;osO=<mrEyOS5bFP(q=!&
zqaQUYi(s|aKA=~n0jX9<nywD&^NAf*JsuumG^e}q++{4fv(e4(Sz&scDi*EI>Q6Z=
zhCZlgE;ZZEZ6pYxJ*851RdcjIjvd~&JDFIQ&(b5_yOo<>9rJn9if?x*#U!tOkXnDP
zH4*h?;XAcYd@f?0L3%FZ=L1w5=(Nk;*Eo;vz3m!s<p`2Pdx`E=_w?+=PVAO*83oLt
z9fuv%$72RK_ttZEokrPiGnW{ilrcDkL*Y3&skdO{7reXhOui?`K(sG8498;PlPehy
z^sd(qn-ASVMFbf|zgs8Zn=|hJwJk@32ex3oyC#&uVtAlwmbJ&mqg#`&UBG@Q&Fu9W
zTV;1qLXWIK)v2g^Zr_wb*tj-^n{=Sb<$)vP!L^HLF}r8_&-TnZxO$C@R@GP5M@Ay+
zWW(}ZG%-c5B@1}DN>aH}MOWMG|KJMp-l%~>pss&T`~hn(JGiV>VehSn<bU75$NOy2
zRm8p`tR717vGD1R&6q6Oil6<Yvc<vMEC054NtEZ0QXAPbZ}Pj+Q&felj77D5Ho;44
z>CMG-aC~|%tqX9}*r%-CofBo0W?WDmpckoKv@qIkkan}^P=vN8&yZF%9V+jLqe~>i
zLau)?gZvU955WZ#Gy{GRZE~(4nBd|-+*_lj!We6g)?4iIbLnql7(6L1akl!68aDs2
zc})`qWwvn|sfT_|!l6nX^8EvSdFUD_b@-p@_${W-S(tpEd;`^Xexz)0ajRWy@P-JD
zTI4$>g?6Rh*H$Kd(zJ$Y3t~!x&^1P{U8u2X@8)Q8`ojt%K?;FdU3AGQF+e_5UA=R}
zR`e18GGU*&D20*V415?Z3#c%JFe}{@ta#pz#oJ96@~P{+*z=BKm)6?P7@GxHUgWy(
z=FGf%+)r3ww9#43SQa@#^!&s~fg{)eDj%n<YvrlP{-oPx%}oT5fp3}bXmGg7z9AZm
zK7)P_-m2y_l4DvMiUz}YZ&V#FBMR#ZTmtye*Wc50EGeL4Y(+G|e)+(?u+zE^;&bZ1
z;aYq*?TIN!^FB+;xz}|!eE?HIctGShTccL_@~ShpHu;QLPo0MMU7E*4JwBf>w;Pg)
zwi^SuG9Tsz3cYvGG_g7BargAoId&V;8tpBzku7~=n^l;6i@HlBpW5Q?@fa`aj>wFV
zee{Z5hpKCjg^L~_B4nv9jNMV5qJIJSwcqxrx7}&k^=&6Km|Xe!QAr%NBARU?oO}x`
zM(e6sty(S^jgj(C*-GojcAW7_>C)l7ltKmFu7_5m^$O^^iG$L;L#%>A9y-`YUuhmG
zjsAcchTU_O2CGP=QVGlGk<K<BAmlG2UyFCKY-#!N;H^wE1ET1qA&0B+Y}Fn49BsWx
z5A!Mbr28WS>p0J!k9E19$lG-&zR-`75{fuKyq#52?jhYqk@!4?VPWn>Y)Mu7U=>F*
z;YZE=6R%}wdG#P6YKBT4dV~u7^@qkwi!Ci?xvWQwsfw45T`CyPY+s$7XH!=KwI+nZ
zE7tE=k>}V}8AE@|WW*;MqRhJUu0xieZ8RberZtSMOgbg)jEYZhI8`+6IJ4w@jQ!-5
zcW7ynQTygi%(DLPaj%p{egu>~&o-74VU%H1Kx^S3O5<PZF^5D8QkR4Pk^4avnl&bx
z2@)e@n^dc7G`6CIxoDMVY1G(O+RAuSfzVAy(S`Ecm;XeZ*E7hMJmOzqlh>~?E!sN@
zz<-WIMZNf<uq?e#DT2XTv?@-T5kuUJ=r4nX{FG8#;+*;Q2h?4ig9YB7;5JUYbyGxZ
zuNvunq9|d_I%Cc?ULt)>!^vSnp0-Ank9yoZae2T+e?Qh8qp=Y~?j-Xto77kqdOd65
znr;&#OrKtvILvSuSc1AvFg}!ZjpDB<ODlZ*Q~m9nA;+4I>GwIX;9#Fw&elcWw5}F$
zs`i>VD`w#8^3jUAd*SmzG1x%Srjh;E^8+@O>Q2j-ytM|!qJ|5C3}IHZB*rlK6=PM1
z7(mr8sY`BpymTN3R`r7LjQ`EVrFixQc|*n`#CB<d9tX%jaJi`Ec7T=;0%wH1K&7D(
zQ2uLsmIfZCgd?C-<BJ!diC;NL83!=#ZC9Ika-etd^l>4!Cy!iEF%erI&Em$}vcY~U
zMyG3wht=_V#LnD@gLmIuY@Y-%O6k_VvV1;OA%5`yrl1_iq*hT3OeNbaE{rUFG}&|p
z;S4axIsi)#54;!z#>~o~MJYsd>Yx&nhZw^_P>Cr#O9D|E1j`_adkPAywait(>cT5x
z3U;<Yu&33pbSDpxy!V)wl;FXvV*9Sm@|D1yj(ppwrEmRv?sbDvvl@N2USe<eJ^YQ0
z&azMCP9z)39_&2*DXFsK)yC?d!!!3b_Sp*^@|i9^GtnPLIGZO)le{tmU1P&xE!}S~
zUnO5E>Uxj2L=3_halN1ih+*2wcbz84p4HW>_lnK`6OD629+KH&5dTCiL=0BJOSVHw
z#USL~3BubI)n6TrgFas6N-|bR3|^^bxppRD`kZA)r|F{hEruNSPGup3OAVKV!kxtD
z>YVds^#xcN6z2GabtrDviO=y1?L|#?x+v9Zs|Tg@9PSzMJ~G;}i6gY4p45M9$=9>G
zJ$`uFwboxCU(dSAte~NO5@ZlmMg2GufvfXCEkV&;Sz5*TV$nqH6E!Qsf1N?1bt@Fu
z^~&>x*bG2RL!7xfP~~~n5X1&IxF1{{Y)-47u*!si2BO2rhq&mKu}aus4)*p@V5{HI
zAcdEwJJ}i+Ex0uKHTD(zTJ7e5ld%77W=@X2QF`oQU7Bs6_s0x3vGms;jFzi^HRYAx
zy|7uiJ-5h1{it|2mmRmM@VNawMvn3ZiUAH8CDRWI$Rt0C5A!<Aw^-SKoeKxL{QHz0
zGdFj)sXOQhK|`i9we<pYnMfETgzRS`qAmzY(*PZ><MIB6dRW~5YwV)4f2bm`Dil6D
zOCZj$Q0K7PTWda<Q_kt7__{zp_tp5##j(IS*4xZ|qvPz=(1eHTRGywrM*dp|9X5mE
zJo~cp8tddTfXOSO*ig?^nia9JX36X@<g}>n*d-&c$p+<KCKwP8K~M`huiUj~r$^~D
z$SZ69Yh>Ge7?z{m<wx*s(MUEr?t@9bDd|U&oxvC0Cuh5>cSW;#tbXmUDXXj{g#UE6
zy-vr>{K5z_pk7-|_8W~G{RI8Wg=+X;xN2Da&h=Q$6Q5(C7fIW**gKhi!i1L=8yB^8
z;*syb3o7+Ud4|0oTk#H*6x{h%hwdeV@<hM$gQEwl(0QgeT1ciy?^Wi-?+h<hzwk&^
ztGrw}qci(a!;yf<y5^QY7M%ocip3y7Fcw|v+p6yV1atqOppSz~5YA+6=WSzSlkvwE
z3kQc6-q8vP2JeMip7E1XQVu;>;`lA;8GK>4;g9NYl>W@hlw;^R{jJ~Eti5+$=edO_
zcRyD+(?IDaaw_4~HbkFs-*K^f`8i^zlI<b&_$v)-1571)|9K^SgFuSKC*s-9)B|-&
zl#(up-)ABP_1U9?&8pY;GJ3zmY6LlqTSl$2tE(Uc0x>W#ed_HTj${t$S+%bjANg@q
zI+~CqsivQk_*+5rjN?RLxwEVT`&FpJ-r3VnhFjs`^zU|_wB?$mMVGD4vHBC&s#&$K
z`~C0<-d%FQ!2JTUBpAR5xWE8!2xeFy4!e8#^5xZRT`>Jz3==)fJaWnDS5Q|c0cW9Z
zo;IYs$tx&W-m2b!w<|}0I~M)^eH&Ox51MFCeT)m{Ko+vMzc*+Uf2ZQam2O;nOp5Xj
zjwVj-iE~aO4~r$qy+3pzmLX%~qXBajn_@BnZ!I?|KX$_V&?9`<n@A+Kjk2@LvhZW>
z;z*%?N?+aij+%kt(JVOpra=Jz09H8zBjW~$R4Pe(;h*oFoiAvUqNAJr@krm@dU$Sl
zcvu(&doVEcg#^Cup7&*Bupzcw5TyIlkh@LK&i)2cp-3-Xn%Y<^sJ>IgPxRRQ2_9oK
zKUdf%(hkGqahoGl+toE{jih_#zh?F$F|q>tFD0#<z45j_(E0ErLQ~7S49nJY)r;<+
zhe!&yTcH1rWuahc{U93idX;+8GUhouP`*BgfB6oFz=3(i0|J$Uv$H!AkvDZM<fiOg
zQS&9omfnAUtKM%X=)ya(?AUENDm#*|N`5~$3QwKh?LB_25;y2gGnwYm^m*G%Sa@e@
zHQLY{lKlwSu`f)CPEF9@C!fD`HnD3vF?4{P%n>W({%bhT&Bjen$ZUY;z7VOTMBr`r
zXD7nMu)1Cw7Sjqz4?gZR^vsv}!h?BzB>Twt{43)Iz6(7$F%MZ3#@WAp5=f~^YtD!#
zc0cDOBeu!9HKyoS^|5hLf^&cMT<75;d(0=!on1Rst!Rp}nqP#XsELW47q4k?D=H?I
zrXDtNFTUxZxy$dtj`|+(U2`;NRPYgeaV0?2`tEuy+9-J{<8EUTpJ2{@Hqvg4zy!QY
zO(~zq`0K}iPd%xPi~Oy8?tz}$o>5>alfy-=T~^kWXyL+5jMK`6q}2r%6^CqPavQ^M
z)zy>de<kAW{jyr074OFI>N>8~ntPg$F;(>F_z`ZY|0A81Wwx=w)#n%JrBwQ3@OA0l
z;z}RMqpnx=;bAtZlF4;)+$_kY@XXHbZFqB3a^h5Qk4=HNQ|ae>5>6F^iA9xl5o<sp
zC=yIk3>WXNqp|HRr5T++?vB@#`@d@}CH|iFr2Uw@{&o9ih3j&Mp|GYE_Q7s`%u7+9
z@$(!yU+mS$nnf;LZ!h#G_Ea-qec+4iu1J)WV-~R(=?&8o?{9AXJnTN)6!%nxHD~Vi
z1aZ)ave}R`CA`}H>_|vPyZ!|=+A`%>$Zy4gQc?CsQ_R%_Cv}JAxf)c?1St-otD8-y
z_qX1QZO!^}b+ZU1CT^76U9WJP>8EtWE}OBTXW_~k-Wp1Hu}l8+;tc~*AMsk@a1T6j
ze_|06HK$9q%Pud`?_*-|F|D(G&O7naX}ws0l^}DDxO>*Hz0l1Fl`7Eq%Gv73${1UA
zt3S@;ZCyO2^Y!EoKSu%g3w!E@8gJlYo<~f-`*15(rSs6gq2Z;VzEjjy@4ARR=laR?
zVgde)x3b>eQEP-=LB`$1PY14dUiC;k+F-6{?xf1(d-=?QeCEN%y=?0K{i`SPxc8s-
zFI7Voj8@Q<2u@rZA79(tEQ{5v@qC^WYO6@$7|!6>?S(9vIRz<`(b03nK6pAWZSRm}
zG^`fr3y<|jt;cSgRJlF2{-f{ZB=G}p)qmw{vXrRsmSpEOJ0`;Z1~GxNNU{9m8<)#M
zl)XJaG_^>w(f)4jqvf6_5EeKR?4-17dEROya41AZ=5=@TYETFfpKIOWiI~2D(=*5Y
zr(|S2?Yu0xoX;i&uKN>@`{xub{u%4Y6ROJt(ulMQ9xgu`9&RNIrUdsqS<eMFLG___
zQAX~AM!a&1qnxwz%I3vEU8@3DUq1RJ-tB8M_}pFT&lS{@uN_bA>I=GKrHjco=KEyf
zYS@(w+GZ6wZFhgy=UHCk8!P`U8k5e|l*zmKJDwsg_Etf+DGv5U$2~0z8}hIHXQrAi
zy%%xAE<eD3ApR329`g{7-mNZolO2VAjUI21b{M(lBS%v8^d_FTRP2_(RZ0q4g6&2#
zU6h6mwgzfu#TQdd!2M~sU@%KJaxP7H#ypJfi>ipF?g&nJJ~kbaYS$dh=3;(&SL#s4
zzGCK9EJx*=Bd_YP=242`-{0`<S?xzkgxjYa?b4QUt#}6Q_BN~aqKoj0Cw|nc90(Lo
zn+qI`U2PddeL*)Ve#`o1u}ukws+FFW)ps6tdMVoot(5b0%KgLHOgYy1evI<-sR@VA
zZb(e&-6X^}Rn)&^MMN`;X4#maQJ&kx3uF=a@;>(09>s(C1^t3>rXPy$TC0|w*i{TV
zFMhOLAG`CRNYq^fcITI5->-+Q{H|mB-!?p6?3xRgEYkD;dG^@#a1+~CGg#37VM-Gv
zjnG)_0i}l=o6?Qi+X*F~M@D;Y^R)3^apc?M-d}HHDY(V=#PantH3fI*ceTjj`06;K
zU5_s>HE5o#yOh?x$CTDj^|m7YY1HB$HZ}Tk^2L%DhXignTEW2V<;<3~>uBhQDN*fe
zbMfWq%<Pb!ACsc=i?ID&zHZ-pcziTSHbBEZpKvCZV0Saugn;TQy)_2||4G;}%ZbXA
z?b2CG=*PQoyXZ%{j;Mcs(D&+pcPr`Esr#b$@cZ&;k@N0DCKXAw0q%l^_U}^PUmq|p
zF!$U&)UM5!H@_}Qmxyr>v*O|Q#Z^|#yys2xSBK!*yCTKrVv5R%9~0V(l@-|0687JF
z$M}YMGE+N=^vVk<=j=|K@ls=4`uD%}y=Fgi*c!a<vaqEx&R9%qQG40^z+#$N+Jq4!
zcs=;liR7)Tjx*8ycDL%|8zS<K9MfBIgZ=PyES2zd>VG~%k*tz-`hUZF)-l?)xjkzr
z_*;R9aN8w2`!kc`1@R{>ozMCmwuMAAQZp)IBYk69Bh;fB>l!^*t-ZUwo+)45z}R5G
zV)!XmD2C@-T8@Dybs$}Obk6(9*!{pj0>(<_IF((`#=L18Co%2%Gg*U|<Mi~Bx0X7h
z^xEM;I?VawLa|4Fg6($3n09CAG4)V9zjH)HOGzm~fwT#Qo#qG|BkEEf6Z{n1*J@oe
zUx^3O$O?BK+*`XKF3$G7de%rz4|{l&{)u5^Q0IzREYw?9@4ch=_jaribrFeI=eX5F
zG`{8hXi6==#p0%3-DNuPf$Fko_Pw5R4X4y>R=0#Y-oO!`ROYc;huAa+CznLWj_3#9
zRJ9n9^cZ2}T*RB_lu3w)cz9ztru=P!tg`~@y8nEHAunCz$D73*wpSVX7?-)mCYL|=
zm2o!oEN$;odSrgqQM1Tl=pz}~@O>?MZ-5*(?0M|8(n7$ic~lDK!@lz)r;FZv?(?oM
zyKcb4&kbR>(Q~`d+}~^@R>xEkW1<wA_(somwa;OR%qSL%mUD}Aj&zPzR!!*=0E+3l
z`dYZARjF-l25WqJDd(B@7_YX%Og+B3pvwN(Snmp2UldO^<|3AtqOhieyY`OiWhrW=
z{KWO~<-MNwQ@m?~HqPFd2j7)w$R88_e%m`i*CT*_$%s>FJ!~Ci&@RH;Vw&68bLf3r
z`y{)Xcb>rT?xH?wV)TUOMMi?U6-!onmd^lYn#kguFY@T3j1n)3zcC>oy?ztbG3T3^
zJ(7=-QrHOD4Tr7sqr&UneI<hQmerYqbS0Dcy{FwD9}emH;Ma@xFQ+^|eO~i?@haAb
zm%~PdAF91;KUlh#nz!iX^k^@fO6Bt5&5w5<yzIL9^_s*JS0=HH(;ta`B0jsPfw~eS
zPMt+lrWkUZ_jmWZq;TnlURq%@42Z?NQX|C}dS71p-F^XJ$a1!JTw$po9bLMM^pPCu
zy0DBI8O>cQER>0L`vaXP@)x@1Q~sP%kx?(7(1bp--2Lg>fT`!P_~`hJ*R#&?AGA8B
zmp=6VN%3v`6t4uU^7rwy908=2EWKc&eMzMArsABB^NZSPW3T>!&FY=z{bpiIyypCo
zVV#NsjTRPv;zZ=h?7TM99AXAU7tkMYjpc67{6;>$uAB6lv3CEJe5!C(11_Usx+-X2
z8b7wQ;yA2U=xSa*-SBoF?W@Rd4ZemkU2@@3j0hfwN7%<knGGm%4+cKy#TB7jb;LrO
z7S_LrKXiIBzL}F~c;#SoDt6j*{UK~i(TOAOpQpz}KmE)G*@OQ=^4r=$(Z|^y;vX=$
z7nD|0uNiDz>B7u1!N}q8CVGr5RT2bOFijILMi6P-I`Z;-?oiSi^g?eVY0uYE^$dLR
zHPILLu-|@c{BG+UcJr68AnzI4tEW>DzkX<Wm1l6)+vD-zJn#Pb8ucxw4Ra*vZ0dI>
z=J!|>pzGqtI!}vfMX_o9IvP+Me|$@Bo?$ZJNR@;kUkJTdrI!qXQiiCN*9bG&uy$Tr
zls`RcKRuWCuKP~bq&$0kDe?J$F0|NmdLEb40&%Zmub}!1te1}-=>I6;^yE;fcrwFB
zNtXAyLb;op=h@?+P62+24;aJY#TYqFBU2lA^!^zY3hOQV%@yXnOfr_PbO|&*-w1-A
z;sCVL8GF9@z|ZsTwQp>5#T`^|XbrxrME$rIUVin3*F<(kyy!2-#CNAyDBBgc*8<LU
z!)8qm9QE&s`7p#jn#9(IW=I`LJN)^=i$O_w`{#*lQ7x0S(skHPfV;hvPbzr>nf-}r
zreGCpggu`W5VaRZ1t}70=o-0<IT#Ak!9GnRD0oVQTnN?yWu@~fG0J$KF>dopb*yaP
zc@N2l+SYm#CTE1Z@3lo%Qd1~4zNF!xTr8FpJ!;mbIrTzI0c!r_ZC~S`?c8ui`Z9qx
zbtz4AJd1A(`ImW6q9aFYL0vatA^!fm5S=C|9uq!N{Yi&HZ=yrpm4nR1ALGS5#f0x4
zG!*3U{d&tJG-hypG8(5=BtD;eM;kZT7wtH-t(}wfVtjt7!u3Z7>e#eN?A^`+yN$t%
zs|Ad;Eo)Tu&K1v7`X8E^cFLfhv~(>0tgMREJ=2{KIWg*_a#YTUUu#UEc+?X&q*bW+
zo&=xXjbM;KF4z0n%cGl8d<i(}VtBP2V}f)@TO)psc#4ndWvFZ?CE*uuZ!sxld9LO1
z2b%$cVU|ZZr+EE&Tp}KO{?(NhKa}1%GAURp=tfos4wlz<Qn#-?7|1BTvy@ZpGP5VU
zZ?_h-{B5%PQ=6W_R++k4!S%75@hS9@qlZR2R4Fm4%CvSuDB6`sJz{#Pu5i6XyxZ{@
zzJIpFg{rV}+BnTzq?e%UsIffFe9CmwU)9=S^SI{(Ssaw!jQ&Ic*%G=(YuaK;YGkVk
zE?i3mx?Qd2o<BZn(9SOVnnx19Ly1Ehr%1^0u)9Ut@W5Ngeq3d+)D7Q5>527D4vn$b
zAgQ5<46Kj1k6N;|!dK}1pT<?=eB<V&%NGtGKVKbY^?78KidfH&apWG{5Ur5Zbr)-g
zMy|3ag*S8h+HWkLbHn`6PtvZp#~L+Tvaldi8__-WvGUQ^%dl6%X;0TNbGQsTe!IS|
zG(D#N5%M7*tRX1vn1RM{hwg=W`@!vetnG=C<LsO__*a`Nf}cajA*y0y#En7kzfNA=
zt88P&nXJ{2U6AB0cHQH8;oX(hPkEfwoK0+sg;H0ch}PIhoDp}@T<xv?KuRcf^-u}J
zfHBwOQ&YwT9vto)HrVB1)d_=tYFn;m)cWe#KGAB)CU*~>(VPiakN5Ly#N)o{=)&&V
zaJi{eFT88lJFsJ_Qh5hpnD#;R!5vK;>;zxj$zbd-mXP18IOwRPzF$S5*mOO3MkN72
z>=?N0Z%@737ShhRW@cv3^-MXF`PN#&$Z5>pucrzbiC%_Z?<B_U5a8n{-SqA>sD7{P
z;;dDFg;X@~fudbT=*My)zA4Q{&qk#0G3n8_VTU3;f!0dV`VzAHRwm<`cbovXLObU0
z`_aN1Ztf1^X+*)u?vyGgrd@0=y=hr_v9?U`;co&;9$HPD<3mSyhnG#GAx+}pt>oyn
zlc}FXZ)xTL`Sn&Oxuex=XV59Js1+3%G4&Vze%XGV*J6kzuC*HZ?Y%oDqR^p{Q#U42
zE?u>?4$65TPZ*4BL#guTgWn}_H@VUA2AvcJZWa!Wt}5s7mk+Rc${i0E3a&k891N`?
z_Yo^N;Jx};;I_-yYDuxEl?WY%oPm+>@*@Hv%+am#;_3&^9}=rs%t`0G95@G$U6j@g
zA2IvS^Xub_2L=!{wf>Tf%X<^`yK2}a4&zcH14i&Xzn{n3F^$pq^#c=h*8JNJypJ*s
zKuc5C3*#BtaaTRqJoo2jXX|Q-!=&wAgkjXO*R#3qgs9;*+}r1Qgq}HO)W5JL2M}q2
zmw{nzb^o0~L_2bJ!eWArV3BTN#3x1XpSgtbd(Z{-=qDA>dxV)*a`T!AQKN0XgbXR|
zP>%opeES_dJO=**w25}pWX$r$u3c5j_b}fHtgQN`hS>x!w}jN`HQuw%uisw5Usjq9
zQ_pBVjV?EAFHn9Xa;a8%a-m>+v{GU7)N%hNe$E90;fX(mMGt2Euv_`Hy-l`(29SrF
zt{v1rDtReY4NhOmvv*8k&lyu~J9;<8S*@d2XLVONnyj4kiv-i<f}T<%kK_^0{Ql^{
zM>Vp2m+4EQqmExkB^Ny!P983DjGPi(NKW0_I`_kkmn&(L2h$Hb+#s$!p?xiFOuuvF
zJ_m+Tj7e$N(0UPUwSaz$+4dPO1L4y+rLMvo-Z&VaP@QtZn3+c}x!gV>7`b1}xZ`g1
z#Kx}bVcL;*b<$S7U=l}VLcM=!`BMvQti^NT)|DeT*Yl5wK6VD{+1m6jRX!0He}5@t
z03j*v76(cdK1du$I8vEhePk}PR2{TOdGq9w1WC%Bi=x$6mJ`k_ETL;0oU2u?P`UG6
z`ukYrv9j}O;~QN(@exTivOV^18AoE)Wpi!LIQ7%LCrRfD5ek6Hkr0B@MaDyJu%U)x
zxWr^ta!Ry5vwKqEC=%MAs0?oXj-g0r-{scSa>AkO=<s>-mh!c6ccfN?w6e7F%j2*z
z{tUjww?J_~qx<Ubl`1+X3cqu`alL3HVQjaagG_7t^DLn#z1=~^14922zQV`%VrO1{
zJ!sun9Mjq0-Y?{{Wn*P+k_e>MxoTW8vG6K`)Dn-eoF~?#*V}sgUF~^ySIcn?)R^e%
z3*|*;lIiQMDaqm5@0qCk-I)QsttN!)qZj*&wx#1{<50}ZnVB`EOC0(8^=xu1i({Yv
zi0IvO!r%+N;usS8baI|4Y>W4ho86HzdDge2{9MfI+c(Ipp^KW2aTE%e<Dyb|^o}d8
zS!?Xbf`rhVD~E<m@EWJvRWj;lN2r1N0Y)xfo@WeQU0yV`sk0v6HgiG&YZ=~|kx1S^
z*Fc{zR{z@z2m+|d&DPU4m?dJ%k`g7?FYsLG;ay=Fnl^rf)=HEvtsQ|HZ(xY=ZT;ue
zu~lb1E<a8&^nbCZb}VP3TlLtCW=E_#r?UPkSTeOm!=_7faHV)Dui1#CkhOqt8V2`7
zWynoL@LsF6X3uyS8^(-nM!LPH&uFyp?%j<&#Y4xw_}>0wQQ$Rj7~C9BuEY{lrzskL
z<f?o;#5jLh#K6A1u->lz$SbI~p(%%ksaP1^=z5}i$LirrnFp(4`;};%9#ZSh;pGV(
zHF++@z28p{X>?E|bF{5B{;nQQt2#e|Kls9OhlOr8%2gg=5V16?<`$cS_fUh;WyxXA
zcgEVv8PAvks?XnhI}j*~^LgH{2!(XnF)IrJ-iFE*m`ui)*`#$97@NCru?P^G5MEIN
z2wpXF>?vPja@*mk>-;ZZ^s3jqt&gm7zw5u0ww-E@G-3F5t?!pB>5WptuRNGNywo4|
z+1m?+**@N1BAPuEU~d%g8=R}_obIJ*O)W3vbf>29aK;czKb`4#=%)3;)%rBZr1yZ9
zGatoJ$vm7O&@+XSvKN<PLs-b#t(#w;kH30_8avY?$&zKOxtrYiiP`SYR`@N_T(kA=
zbR%^tMMq8@o<p=GNKaX&%C{%?*Fw+Et<SttsT$N7gwy4M`0nM^NU=VVH%$w6IHCFJ
zj?hmTUsIcmR5QNngVpC5oXU(Jd$InFO4mV@+QW~`r7@eo+HhRA_hebhW8AGWjeFHe
z8xOC0-hiI3DN9H%vJ!h<oG=(!r${BgM03{xE0R=#Ij78F^sG$PV#Q1e{~}Yx%n4_~
z8%CinD4hfssANh%zK<=d`)Zm)i}@>LO2=;ZOYv1tWw@(FUPW*U-wj<|pRtq1Sz#NI
zg^>%&)Tf$0{fCc>_6ehI>lMn|;Y6LsJ_aole#+e&O)0eddu874Zo$TL`9A$2RK%Gd
zqQ?{Ksl%y=rv!ljq-l8kb7L`@^cC){Vy=ugvh^dcM$T(}tZCyS;5@^+k71+_ca2b5
zt-}6mG|D6KSa7z?o=HmiS6JlP(DD|O6kl#Vms=o*;K*hUo*F47nJ}(T6Jt=VV4A43
zmKINY2`bG*yj~^@DC{wjDVtNC6E5*d!^I)P@BzxfdRf+y$@jb1bfPk};!nh-5M1Z>
zCR6-EfiX867*nHZ>=Onb7FXz<>b_mF+qi3F4A&CMiyeQjP>|BcF3B!2>BQr8w7Fr@
z9R<n8L^Fr{7#~qVOP*K0uYAg|at1w#dz3B@l2vor7*tizz`5pA(o-V3vt?mT6n<!U
zMpNlnK~Mlxqc&{IZDtJ||IMMe8>szJbnP=`j_u9hsxL*+MTEz{$2WqR^a#jT6YT`>
zeDRQ7uitOd9KByD;_Y7imOy%MN;6^A=ec?x!y(SDY{ooud9;Qq$ANz0F9q?i^;b`e
zG@L!eh25l8(reK*ZNe%91qQHk)F1ZRGZ<83S&9Gj!M-O8)E2=AL9n9)MIKVpp4+^O
z_$E00sOYYijh8pC713_vC!3pN7^L|Ftfrn`A~UNSyTMYVoK!FH#Ez$MT4Y0Q`!`<t
z;aqxvpWJ5LpX^^SUq?trot*Xw3)<lujWW=gzS`+Udzewhn3tGYM5x(1a`i4M2AqTW
z1%Pg6@t-6KsSe)E=cv~(uLGL2LLSC7X*FHP2FjM53#pksE&aCEGaZveJNa2Jq<xEx
zGjrbTOf{>%Us+LmrL1~JRL9=5NM$(dd{avKOJ;A(5F<9@InSkYgC7hp7rt*hyAxv-
zXJ~>?Y;?I$|L)OL4+};xdjrYeeGzf|16#D#Nr>Y9o2Dtv19$*qdln>G+Ck+(1Fi&2
z72BA&SI?zWt#>!8YW(^8ZS@%M=1&*ryelfU#M4M(WJ~xe-GHx6wmlyV$YCJrS6-JE
z%{6G?)~^9f2s7hebUpq3{CGA#V)g6x1rw9@h(_aWqGVfU!vn#6Y4J|w{bM0Jxe0NV
zIc2G?@sa8$t8VmKTI0YYr7~5nIUD`jO;;JJT^M5W=U8}+k;*3Lz(MpZ4|t~y-ouBg
z1pWv9rAwvhq*YJZkgxm1j6c{x`6Wm|>Bf()uZhd*ovvnlIYis3HQ_%o?3yzktoufp
zsVA0GP*RItiR_euLr%6W)W!WyD%HOh6cEG{V9EDRJU2*R*xDNzmb>9;(;8l%TlJf0
z!`J16vCdiIp^35B&JJOhkLEA~K=4q2;5v66KA@bO6p1W4@ng;}dEPO4b$>1UW$j$=
zY=yjC_o%+>MBZA%B1MhJ^==HN+V7s+Y%oFmV0F+By~ew3roEGVl_B1C-Xftii>EsF
zcc@oqrB9AWV=(`neN{Pu19^|N`d-|#ohHEjG89}CPwXI3ZDIhM*8o{BZ-*YxidMse
zR9Xzu2=G3h6kR=F$l7yMqV-~mKYdeHf6OwpQR&=zLr7{<+%2$WQ%rp6=(qVc)Jik*
zoY%8je;zuI0BH~Gc-f*LEX#9BzL}Y6(Vv_)aI0~kH8pS$7kc?`a*0%Zx642}ig8_O
z@11+wn5?ny<xWmfQx=0+S<B$+KEiS9gwy(L^e&xKDSYKf2kclrohRZ|Z17~Me=f~=
z`;;9QbN!)md6{!g8(1<R8@0OB$kJ6fnzBnEHA(}k#1l{397<)kHk3+ZSv9i!#Kdp1
z0!iW(^{+htFUH<7tg7aH1KofK($XNQ($dn6gn)o_x6<9+f+B)QOG-CLcO%`s>F!N;
zo!Rg2f4-bg$4e2V_L{Y3=6UWKzR$_WV~2M>y%c=~Hj5Pc-6=P3!v(R))UVdjdgAqQ
zg`x^|oGdLEk<}I&67|$Q2rutlDO<81HxK@}Pehf!T36dJnjj&0!G2NOtw&TFDo69Y
ziK&VUVIi9}cGZPh>6GWDMj$ft6&!|uKUSq&E~4}z^FrTMq@|493L1}ZjW&f_e%V!j
zzSy6~Nx^zJqWqfnbdu(%<$L;@>A6f?8+?E$*dH0f5rsy91XF;LQRrS%^^eC_*sqC`
z4e6`y)+3kfYM8Wl8?tLgcr~)>Eqkiy01*C0`42<`8^`WV3*r&`l#94qfkJ7x*x4KS
zC@}N)w7;1Cu$i)#`;&N6PG1Z2(&u+)_M_JD;=ZyScIkb%u>EW|4nL~w*h;2ctM&?*
zu-uOEYX|?nZNotc0L4QNRHmx5teOO04m)i4ku}`)W&iI-d->O5pcdU<p9Ro3!YHT?
z-isb$l#t*MR1Lqr{~i$qmg#47*x%Q;ZWUMCt~<(zU93MaukZVnZn7>N<YX0uvh<&z
zk9!7Rg^0Q0dV6xcf7VIySzPr}LBb%dDYeG68QHuE5hAPz6X?R`-g=wr4^tLF$;url
zfHBN;6v~tuJDnx_UCwh-6&Lw${P&?dZ)pDl@&fubC4?{tgdYDC#2MuWqafU2J>s<0
z8>IEnN;X7>_oE~K1O`=A@UTTrvk?v)pGA5U{el_TZ{-zCrdu)?JQ3sNrJ4~-D`}Zl
zbjUn2cQB<uq)v1)a~Sz2!w~>;n!HZbkU|diD4o&U_`A=sz}FJ>MSql2*wH()`}Gsr
z^z@IG%r0~gH@9J1&p0?)w0?wZhn+nMO+vz@#EDcGRWd51YGG_)3YoHT&U#{JHp0N~
zjZxKKwjRVbiUxndbSyKW(<RBBC_D-v*M$<9K%L^o7{$-s2D9a$f6pT=rb{uPumw>C
zLZ*-W*Z?UAYZ77{MPq+p2Pm2e;C}h*`{z0wRMcYWrUb=&)r66Po}P@vgc1`A{`?k&
z=a7fhEOY&D?%lp)$z3EUjpk{jz^beH^>|#^0rsrlTUCvwj*Vm_a4|I<s(*k*0sW0+
zJeD+HK`XSfHU9j-Mtml5KHL2>z&ZoLIyU^9>vbUIq0qPPA+m@xnc<6Y+!8-GysgIa
z-)XKJXGJ7Y!FVgb?b>iLq`6yfx5OncONw5<E%BvVV8C9_^r8IAVL}6_Uo%C|sr%kd
z$#<-X%Y{e=E*2d19XrdJO_aymhmp$4Dr28Z$CKkuM0Dq4%REJ2dIEIiGk5XpYe1mr
zQ^@bt3eR40UEx||*`A)>$};Fr;}%s9;kwW=bTL820kxZ(h{12mXMxiq#2c!kK>9UO
zX}DxS+4gpcaL9eft$8W+OJ?qqm-n=7E^@6rWaTv-4Q66m5jD~@+RF+OhqU*i7euhF
z>%n29qt<m4VnY6b`<eV_xwz(NuM1y(cbVO*Uq4roQ?7g1+S96apcWy!35dM(nWrfo
z$8~#<c3uO8un@!Zd%*f$K0IXH3uS04nQGY}dWw>TWmH~Y`}|Q(95iwo#;r45ZxjRF
zM>PQbQK>ju+fB?O3<qE3+Joo!i08#jn$EukO1rfQzvA_NW|A55g~v#)(ae<_N@}X7
zCztI4W}<(cEnq`{Q~~&fi5s+uKTcMM2D`))Cu#g&vaeIE_tgd_MGf(rWPxp6i15-<
zUGFwlZqTJ`OSj~bL1%TN4J>$yKw8~-cdt@P|1-tT=N)=_Sel}Wr!r@I%qsq+@0a5C
zl=+2}y5Bg}m(e~N{xeBlRREMq&oCa<0n!6c8?ur_WzdimkD4L<)G+Xg1<v?4gy<ju
z(ge7*<tZ9Kf;CIykvIrhQNsTMKPtXs3x7c+FYO!37Pi@7N3Q*2;~xOx^2HdmA*I=E
z#cXMu^!{^Zv~2W>n#uV3M;ixso)Qh)LhOB8TU;7lBYe?B3`YRumJEyR#ja-R!r2?J
zG2^Y^wW)BXTAtyVO>-2*k}&la4(!ED3#mGOweqbG+2R(F{*f7iXQH^?!71{@(5(%o
z;FU7vxZ#I^O|Ib2kUneTHHMpZgMhj`gt7x4Pn%-@jj?FN&I?ComG2kdZKvNTnMIif
zXE5J<f}DVqnQ~wyPxU^LskU9{1EGNlB7VXi9zPzV3X6(9fgtBM+Ld3ox3?Aea<>=Z
z7?FS-Y9Xa?VyFV86u9)Wc3FyROS-j(`WSS5oemMd#qM1s>HKonj(^sFkd|-KsI1vy
z$VLNyHl{{7L#MZ0JpV~r>WwyOMO<!8q+Or1#?#ZoF-sh|_nF<8XcQ9`|J_zVs&sTM
z=?wrb5GW0<_rCdkK5py^|3=&~V_$rGRS>R5?wL~U$lPMd@QAd_V$a#fQO4KcW}yRS
z50w)R<Tk9{kK(;o5>9<BVkaQhLCc-ORb<;WCy%P9D_2&quZ%2LS)}+B9&mvO;u<(X
zn5q!)(tiSh4p%ofC*TZG9?!cN29+ytENCxC4h@;LhQ&CArBb&VuSf6&Y26?7-t81U
zKf^xnf~n5_)*mQrAzv@3;byxi6>&OVt!gZD76CkXHgrlJ@mD&p4ee`3{Fp89n2W3b
z0&|merGDgGo#H*%ul@9HH6XU6PI%+SIb2?iR)p&SBbrN*wlBM)(s*>|Vzu@_C>-w#
zuO;#JS)A5oTSvUitpyxCQmR47hODR^7yzn19yCyYyH3Jbh~~)o)N%zNiD_(<+a%w%
z!J4;?Yx?X}E3to=thR-=8CLO}{h+@)=OYpv_)iih@&{tDYU|l{5K=sUx#P?aGMc*U
zT@K_5tPlCDEG_AonQh9x!Fs1ZE$3(<X`PSs-aU-;3VMxdvLxV0fB1=<^zO5Y<#X1@
z-|Yl;)`t5A@s2xwQ^x=;FZwW)TM7N17Iq`u?jj<(FOkO>d^z-`vP6c0@L;Qtb0)54
zaP4#Yfr0>bsb>O*4v9%ex8v>AoR0Spe1S4V8~HU_0lm1*Dqyrb@c#b2r6mDyuegR7
z%R3~Br#JJ{-!80pI|Lij2>pwO^$-6ls-YL4@v1ta=IxD-fu6!sBE!j5k~{W77J#C;
zw5e$3$SCy<7Z(r*P<@qr6lK*5nIs1qA9#OYa&!R^JlG!1<R!xbXGxOIQxXt#HTxXg
zzRhde(yg_p_PgIRm^=U7?gi+oZ`kJ>zj}}J3S>QvX!fuO%U~Z7z+XQ#)H^~tzF70>
zqWWjF9&L-}ybfXGV~^>5M&0@J*4?(X`H%Lu#YmPfXH`dS*Y?K^Z2K6)Kg8Zzff_uV
z|JAkH^`{s!97>#}>{1Pj!gc#zBfwf;37xwAI*C5wxj7w>88)0C(eZ9(BDwEqqCD|9
z*#aob=<c+BqYrzjjUT3e^i4`99-W|AjDc!qHx?#|Y>~kmEds5gVeR4Gdqvyxy~_AP
zvXkZIwkJA)I_JqiYjZ&eJDBpIDw&P!Y?S~aKE%rOtgM{Qrq_iH4EYQW@tCllE*ImE
zF98~`JtdMB(Nl8di;eSBm@32OMi(-cA_z`Ie4|;`dhcRk!Ek)6=e0u3AD=-F0Nf~?
z&F;@Cx0wZup%Ln3$M)1oD&J?4)!%E2(RfwJ%_iyf-R@P=KpfhHKggm|v(HDO`r#Qp
zn6}>!245$kUh?3Xvpd^z{Gq`qxxp1~!NR)3NYj0`_3(mLch;kn8ai0S!^vokFLNzp
zJ-7h10IGG-HLyov#gPGlQkzsE<^6TANR36xz2rNVRWMrRK=*RObo%Y+hxsvIOV<4=
zmM}%?37bYlZTakylcZ7txkwV&_XhgHV=11A&TqJ7YU}%Ma8Q-q`nAJkMJdSy6Cf%O
zKv#c?GyA0*GZ3@QpiiN<rrVPHCvJo`Nvv8*G2zcHsvFkV`1<q)IJx*8Os)Tq{xtg8
zrk;yJ#q9hig>T!5_>x8+gGGW~-qm?q`aw=)3ulwnJ9AW->1U($WKA6i61(JC8Q6CT
z+V2K}QOpr!f?E>5LO7!kOkX0tk$5__uw`bu@%WF{pmzV=6^vr$;kmaLiYmnylFNE*
z2VS+l!UN(nv}W2S4W@!KF1!rdpD}6hT{4;Hsu3&A4(OO*59csiyv%pZe=x8p27vr#
zAUtVjsK&gwjXEe1hc5ugVXkO5Uf5E2IaPEYGR=3d6;h%s7d+Q+nCvOkKIeB@-Ks^x
z$=#L6bw21@sb2T1JqwoAWRP?LV$`yqMqr@IOYIG6utlLd2>sbStZksvTUR%qrdXhP
zrU!aCtFt;t=O+$!5b7MqYA|%Ym4@v<zVqq&WMkfCT5n-9IpgvxMy35`0`U-8rhKHv
z#EcBT^1xK(OGnv4K5hMXVs(=~I_f6fn;&}_ifMBz>$^^G6VD``BplZev%NS=z^sSL
zN6u>6Ev3IJRiRe_8!`^vX_#<;LF#fv?A%qnCrvgN7UQnk8OICf;G0C$V=jvtUs`dW
zA5fbELUMCz{vL7FbmYwD;p`Zf0#JQ=gu3w;zq*M>Y3|&td=8o(^PZ1Skuk>*BbEce
z2+n_sEv&9J&zq`P>qjzhTuXS7gn<>IPQC#%ILztO?+TwG5oEg#P;dDclgDsHa?cAG
zs0C>^PFmv+q1_8HzkcHaBn6_a0mh0h#<N%i2~qpoib3QZv>0zx;5T5?AFg9^G72n9
zZwzT#7%3YpyVHrk9$9!0mNNko0LYHvJK8WZ6$G8|%`%AI|9rCE3(oq81%*mfSQuIf
zI3fm8QnyLEy1J?f7PNGl#KYN&?j~SOkO89Lhld9Y3c*^nE&TSbVQa&G%l4)iSaxT}
zmPtXt#gDwfMaEpZVxAa#cR~uD)Lzz7>XxFA#&YlvtGsdxv3HVUbHWZcU!CeCOOFzc
zO<4Z={#bN~3AN!-c}xo?ZHV6gWmzt#jTPvO25!)|XD8I$Gvdzat}pP`-+AX@BfYvi
zwmbMBPst^0ESS!W2Q+*8AK$S9!QbxzF36qPq8~Ak>u`p+Cy(PQm;g={{%`@-i|${F
z`#d|4HNeS|<S;TI`}A<Vs33MOe$->z{W0)0rERP-yQoKVG#jC3YQ8ux@AXq!T-f;4
zW7-cOUxcLR<$PV;T~*(Ta1R4uEUO49nX&)~(Oy}cYcl&iK8oAOa*uQi1m=KiV#s4L
zA0Mu60PKI*$#?P6aW9uQaK<`AQUAHF=T7$e))xF+pP$ju9U!Ck8WdLZGV41#JKN4K
zY3tzB2X-TIJ)U`xCmaF)Gh1O+0b+#IQ&hfAl@p(LBH!Xnhe@2DkP2LX*8Zq$RHb6#
zq-^5$(ZtO%X)s`NrKoA?TgOi4ur*by@#|KOIC4G{l8OyQ-Mq&X&Zocg`5)yG5d+Hv
zQt;I^84AjjJ}SLa%8qSEdm%}nj>c4sucpaBU-L^(TT(%Wi#t@-)&jFGo9i{#sJjqt
zPy6zF8xLsud%vgBSDd_KY4(09r8g4gO@Yj97-*!OZNGwEZ_vOc@KB!;`MRXMN)5PL
z>}%Qc6yClNpk>~|5DbmH@af#@O`|08&nxH~-%F!q5cvi%fxs!TfSgwDmZ9V;K5dBO
zCuBFp+J<N;zs3vD^dGb{L0Nx;+omk1LSjd<gq&&BGH~(%s7ry+{n9aK<xdBKjcMm*
znbAazu6OG=KNNYPD~sp(9+8}s$5T)$UzR|!GQ7?p2)%$t7^g$R@du}Z6!vZ?m87@R
zu(SVF%Vtj!$i>lXTe5ClvNja0H4qrKN5LRLO*06!gK~Jc)Z+vAIogiCnHK3@`>$;G
zjw+wF(mPiNmxFK_f?1l$PXtyq7RUr-%r*69D!wB2m2Ia%-r*QwN3ZyMR)^f4VUPSe
z>9C4Ed4)Kj87}WvzvWyoldQ0hMbMzq)t%S<Y1>1tomNVI-LO||%Cz|4BV^Tbo~L1H
zpJh0))_hyiyYetoDtc68<`e;_x`F_bKEJ;;Z*@FpsXqS(uu<z7pUV3x>i3QN$^9#-
z5f*6J*`mO3Va1b~daqw8P-t}nr5GkE>GDI2;6K(7ut5RkI94o~<M3B!x)PRc2Ly6f
zRskw=e*wr&2_Nf0l%76Viu;8`;W+r**(~VPl&ofUsy=(NhP--F8S+KLC!dSM8aYLQ
znxj?xlm8b7d3-y@l1fLQ8jNYIKQQz~mJVZN$S?oP)l}Ox?SQP~nv0k8t_x;w>||Lp
zBqdDAM&kS=0rQC_#+6Rx)>F;!Cka)hKJrg7_`&yVOII0b;FfIQ+9q%g3!F{wnEuo{
zoz>l))m`ISL3|Fo9ek>nJFuEe=Eni&?#nAb3-X_jx-#c~RGVC=J<Rf@b8-K`OoGc1
zeEfV(rcZGd2sj<b6Otgz&mOLsL4Vub>u2rkp>Bn&6G%eIo(fceaFtYlsKFwGN$DY|
zr`&`kbU+^tRtc@Nre+=549cWO1nstPBkB&g+|uXEZM`03RW*g&R$S8nR6Q+b4~xux
z!sMgOW}>fw01-&(a1d`4gZBIaAN3&DUMqth_Pt&>B2_91l>1i3%Hs6VxTKq0s&iyy
zeWlAsqi8361q*zv(t6jd8`=-wCt+EQ{}eoaG)<OTR0%MY=`iG#x%OR-kToe#&vd^p
zD~<Xng4fneD{hH*@lN-)8Jw;n#4jnPBr4W`s)|2<nK<$1iUg59k7ap{oQdQ)$KpL{
zw?g}GOwT|ogN{=DuJU&@p*?SgJF_?)fS6kEp(%(P9q1=IiFys}fua;>E!Wy&vBCA-
z868p9T1xCO>T{VX+>upp+I;ScA^@T&FnO1zAIWofKbg)CtdYK~4r#zu=lLMvhD-%e
zf38|T_icMOh$97gY2#EPD5w7N*=!T|H26|<?-pmf>v2*ovb3Qun-m=cvjA|P2}nhP
zM<jj(zld*tWC_f?txEiP{pg?1W3_<a;NO*J_Q{&+l}XYNI<O>^lL|=pTjo{rvuzV#
z@RLh5ajO_j?6Ga*W8Bo=7T72N@Kbw4*(SH>pMq9ZwSm4#W)V~&dorLSf7RZyzdE_9
z1}hgY6fFXp7zcJ6%fcYpq>E(6!XU<^AlX8s!i8t%8{g3xD-E{{eI>gWqZ9rR2h|*%
z*-@%Q&(&LU&^9TcOL@!BAjJhe4~%(!=jFjq7*tI*Se1D0a+?G7@%tVC2&3l@o36}y
z0?fy6Q)jO*0GA9`eKXDJ{~1vK2gI{?JHa9@efuL|scS6TKnB`npaF<z>22xQa@KNm
zfA+$W8_**j)=-lJ{?G=Z_dIvO3zNZR4;ZM<hZ4@y`_h#L&V|2Cs<Ue1R$ZDD7j4B+
z;rk1a_ft!xEI-(vI$m8U%$zdcT~U3q4tCN<RPj3fzz_6{h|E%m%*8FJ>@UuqTs!+H
z7cZ=2LBX>IRd9#}djL~cEO(eC-uZXS%kNA!V1vm2E1;xCpoUC9kXwMP$q2;=Ehw1C
zQ=vCQ-&|ghm}Pw@tta~NQM|ctH@$2uY!cSFYi}7l+Glz6*SgCv_!S9~yU%nRx1XYs
zFKq@Qvjp<*Z)lY`J5cO0oJ|9bz}Y3;={Muu%N?^pLz{q{J?$xww$H-?{*J5ZWIv;^
zB>p?$AXkJ$oCHyA3sS3!DM#&9ixkI2PyoTrB9M9hiz!#00B61BocL@21kI`L6f(`G
zT-2j!c0Z+})`t97RWd2gCH9x5!?u9T;LD`ZLFJO1ow8aiSRK>72K;LsMfVlFYM8C5
z>LX#ttXVi2w`?e_U59Nq(}p`~Mq67yE6`c)A?=~=Ud{_)Vx=7hr5#JwRQx?-3|p0a
zW*l4Nk#N>^UJc>H4+~F{<bnd#Su*W(6(gF=b$V<YS!v{w7k*E<t#G(pa|Pc2g_6&Z
z6<ky;$ozQMz+7A&WuP83x#B#$zTa!>n=I9|3VV*XYu3JA7<=QAhZP={`2AL>Ex?-o
zjOo?D(ucCh=P4y7M72&ggDXu!hX7RkoQq7wYUA?_NC=-k#aypfny$JPRPt5QJR<~q
zBPf09<xV|yfUbr+7n3q**ueu4OisWL#c0d*=J)Inre!oZt7GHDv&j)ZNcr{IY-Gjf
zH1t2e`~>u;ssaT{ANfBgYD-@e?=teJLz$k9p?EmF&1e7$5>y94$$%X<@%<-Pcb?3w
z>n(I0*QM<}uqyS&8o;d%0BLopO0%g2bj%b9M^ecWZv_(+F7~ya1dIwmR<w3JJ{NaZ
zT)-$9%k3yvxM#zKX^YNOgs-NZUuhW-ZZ#!o(n{yc=GhVyW2)P^bQuHLTiBZ3tMsQC
z7?wzAnd;<V=?tBL{u+0nhT1qy4w-fjZ^B1}(S!giH?zYB8!!x4g4a%v2j})Q{<Ckz
z3Zqdt39SH?%!<o$O@ee*-B}urw5~B`B%aYeeTog}`~TK-4T!Q%U`?iws4i%^0DtF7
zvCz9R*3tA5(pxejO!M{V+cqI5zf{We=%L_Ai|jFVnV%pYVDv2zMdHZI@jkcJLpGR)
zSQM^QSQ*~v0kv6ryq|D=-(BamcghJDXeP&UEH#-p)e#R0eY5GqytqvyKAbl37ZgQ6
z7unP=ZH&HvvbhCBu)J3jhahjLN2pJH{Ain<)fdu{U;iEb=iFIc$K3~QL)`Uu&OedA
zD;08>a__S3zAM|lV<=(RjIHV?CKSY~I8JDc^Q>dm9+K0;3U6M-2w$+++BsK5vR_O;
z<<EX=W&AON&_5=JBWadg*ud;P1AjB3x=+Wu&}h)Ef-y>-fBd9~RjifL^aQ<9;tW73
z7rzB@H1OJC_`ElyK7Xe)X|N?K-<MDY;|!vI9qT(|c|<R|=PB|+F0XErw0q}&nS1Y*
zna{2QWWGw8dOdOT%!5OLwG)TF!gRh0@e#vdBk#@{)UO13zO20VzEMXeWCCrRUwX>&
zrb+}_U22~QHKMdsCBICr=p$CY%FG&d(lSOd!H`{@#9T5NwDVDW>yKk#V&g}`sn*=$
zI1&}EwKYuD@%gTrHPYU%LSQ(%E-@j;xm#B5N9=O39caEF0NK8rNzl%eL&VX@F1OXB
zPZt7s=K}R(M;B1IImrK2ZO3H>v^(l3M-zg=?TfP{w553j3#(-jy0+`mj^(;V3cb=b
z?U3`yfYK|z^DT8`D!5kEmWgfxw{QDsl!Ey49!L*;X@v((0ycqcHCK)>riB0*z(#So
z(s8+>aB+jTbcxkFuM1_KgdEVdU7{7(U|Cnyy(!U+zwux_?<ca7M6k}nH(lo0mcu8H
z`82Rq+A!{4`E>1=bLZLdcwD#z=Q~sJ`*4_!jj7aOY5@BS7vdBp$UYPAdj%PnAd+~%
z@;=cH*PG%U2Tp`<G}0vz+a(nA3jvcxG&SN$w0lJ~njeX%F5*aj!c-CZB^iOj>c1=Y
zd^P-~C+cW+@@I|o-V2Tg*@Q>8BMZqn=TXa3yv<W^#hP)=;nKN6P-C_1n?aO@`dbf;
zkN97x3|@&fi-m{TcpkVXy0FptoIK@PKWdM47~Is(D^S;-wRBVd>}1~G-e=`*v9-3(
zT!m{yHWVzj<xn}lsZXJ<Y1Fm(F~5vIuS`8kX`)`a2f{B$?Tch+rjJJVMt3piP^5GE
zy@bEk+})tdEx<DFn4jDbe>5i^F+U$C<=A%C4Ce=Bk7rZ#gferPrRCF3wP#G3f|=v}
zFKRRZn#agU*z|=>PBS-UTJ71Meea5`#(AMoy$UuCkp@i>qu1VJ%X6P--~|3!{frWW
zhdF3Nu_L7G=NWa@o1lu+(C1rV#c%o-lfVEh2MhgWIb9#DPBZ~idret!>k>I1R042E
zL*=vA+@I_WK}`gV4^Y}jHGJeu{XK#;Gj1H19W9d&r<dHc@kflit#CiK*I7oNc{G+1
z+Z}BfG?i?8<;1?kwuh0TF;$NBO>!4whkJ$mgJnS%BV|R+=(`Q`{@IH7{@a?d(x1lo
zVdYB8SJN1D_>nw?52Yo*RW&qG>v1Nq%dpjYV$Es{PkHVwXXkKmBuhA>8A564eaheG
zc$K$4@|xk@DCOPyl}zTGy0lGLnSV-H6{T5Voc+6T1}>&Q;YJ)Zw9nNXXa(q~m(LC>
zoxI+;?FPbDQxT$QG&7KK{ZXl*L+dDJZvvJ4Mm$6C$@~>d1EaP&5+9#DzPF1k>vm%9
zxB<}oUGj}n1ws#4D}ikxP|56*=e{#`#F&6Wgh(Rh><i7rzvZ)#b<NmnS>WwNZRN@=
z;bXb-5AiX>?U-A<HHDe6{?v%=*0{tnvhJKAjCA8ylP$1bkDvZ&pl&Cyx%Z24&Y`XU
zZ~P{s`5>naR>1+5YIJpYuNK~yK?Ls5sTe$}rZSj^@cm1+C-_H8n1z1{_~!0}c>bx*
z5l@(p7)@AJa0p8Yf!~Cg`u(p8?x9U~V<uyJ?@MII6pXPqP6wqcjS*ppNS@C8_WW)6
z1-z=o-|{;0&nF!UN_+^KN=hv9!PcLrf?6TrRLP&T!O%dA$NfA=Uo6)g^TT;qpnmDX
zuCuwp9=>x)w=-Ipc+>q_(5qaPY^eTpStWEc3z~D#08@dO*r<Fou^D}msUO;#ZaB8b
zPFm?uXf70#Y<Y0GS@c*XE0sv-pAgH&Xt{qyd``~dvY-UB-7bG3R|&-K)DpMnX-{j4
zL)3=bcZrOKdF<V3k{>X_o3i@(^DYT;X;S|I!0D3~l!GGaLNmI*Or2Z{xfP&~S0qj2
zY-7=;mq~*eKfcfWZixN$`<HtBtu6qAX9;G3c+z36cCP-D^*6C^Km!n^{To^U0cSPy
zv-`3KUm>%-WlUG;J~^2RYW#~g^nxqOtAzK9WfHfJijqlgUb~c$4RQs6$E#Vry2z{T
zmw||Z3<W92fB8tCzS=!zYW}EJ#vg&LU4%uaj$;03@j|7OVqm50@@oJJ&!PG0Y0?@V
z<<~ZmBoXFo0v!j*(BUmKu=S!tiZ6zqEo)~LzBWwY)^Y7+x)qy5qXZNF?!!~>56}<s
zk-h!~O_1T~CcD<|ZGNeSVNF(S8Jqo3J-O0x(|X;C?@v<*1xcLwt+|V1&ZAaeVo?^0
zA~;taVw?Nn;!Em$$fJk;Y8y%x!V>~k7V)nU?AnGfJZ-<9c89^qDM#nbWqatnw+1*9
z9x_=zgA?2NU=3g>U@&%qnJRqc`!l+x{;3G!`FO$<?~8Cx3=m&H8>*X?Y!(HWRb3^6
zW_)q!R2UK|?^Pr=UO5y)oY3<x`A4y=v#-#Oi~igLDG_#40$q;C=#B}WY6eYuBnGAI
zhmYe+hv}7xZ#0hyo*N;1dme;yw?t?Q>=`=cRbjHG%#gVx71E2|Yxy%4W-bo+N6RLq
zSb~0~qc^9I(UF|(_1l25^sk;j-c}9%DX6NzN8^`lpmH5F-o`%&)X1ghRV8TX_}1iV
zjg=d1tcRhtPe+xGnz8_dh0epNyvgULs;X6^{(G8OxnY@6x@3?Fv1!AcZsUoD_^LY6
z!1|>$=w$({No90Wzi7I}#m6ozEp9GTyF;va4<Tc)R`bu=U$usWOgXkMt$T$RCu15+
zzlA?)eB(1_xHwVbWh5Pvpb+pf9h)BXPsHSRJ4R8lIYy}vW11u<F*I$n=%~MVfuHhU
z`?B?w1cm4Esg-_U>~L#9_R;P$wp%e`2bN&0qYG3ovY??6LlSl1<&u$qCgUOFKb2Qn
znt&$~3Lk$8tyyDuPcGO5vmh%7^0?osDpZuUnpAYQ9SZdlS+6=MML2TW9UdIAuhNjC
zao5xRw#^{PFq3k@9SJ1O7@o*M6yDpKPVfEuee<YlATW{ZCz}^`b%?q;-qj-UxznYF
zH}-U;Ln+hye@j|r(Pc+$_n5C8kr^VOr3jbd0H?51BdEOaG#(^vRtPBUP8kZ~NtLNu
z5X)9W-)AhwVk~wA)bgZ4Rq=&`&H}s3t!sg`g0-tzkj&oG+bZ-2Rn>j0D(<{-MB3@*
zL0Hz<@^>V*7d8n4PgT{JV$%z1<74vusdoj=0%V90-IStE*z#QDMdt*(;)pky5Z+_I
z1dv0<>Glu%G0Lh>XLg5y;XjW0@t}nKs3FN{L0e+0AJ|+6=|-9c;nk?evRqs)^VYFN
zB&C}}Ex!{NuuucGCk4v+ry*&<j@u6A9X**?2)cG+mu@%q;rt#WzyrkPveGe#JTnMl
z#a8sGmFu4Tt&g`vd6z2rf(Rj9;|z2a!1h?I@|(JTg4dE47_ZFvAUi}O*x(BQx27zo
z+_XqDuclv}=C?JRTAhtwDp0n2rsN!6P;i%eEdZ&w85GY|IpQKY;_~pzDy2OzIlds!
z<!25PNnvXlH32_;t9j`cq}Z@yskCGjJ81Zysm21Q824rP$AzNbg=97AmJN%(2V>%X
z6_+i3qn;<@xZ=*uUybZL6wA(qX0B<-w6MRR7=ee@F|G~q8E2*8r=ENqZp{<P-)uCV
zXw;iHAg5v`rede|^=Y&6{jMeuA_13Pc3J(;;yRr~LQc%Wjy^z)6b<PP?hZ`CNFedZ
z(S$e$(^Z3<KrpqFy~CRvu~t7*+Nar-Hho9p&m^rYlUNlFJo4Rrub2e5Ai&T|oVyDb
z9UE8CrDT|BZnITy*q9iMHY;F@0()$p;RGe+Yl^~oZ3PE;_K`X`2hFKUvO%IruKvMO
z1y~S$0S~ufKB?r5A^aiY_x{1k+vS45NmY`?)m`IM{mZwIpundNW$RRw`#zC!AqK&v
z`5Q~vqh4o?#>c>TFP0nH%hP%;xBREWe&^9}%mrr#dw<g2p6~9CzLA4OpA_9-0WE1U
zc>iG8TqksIaUh+OjYrJt;tEU~Oo|FV_@!ey`&^#Fq)Nvd;+32Ju3-Xl$kx=CBRVN=
zu@i(g&}8!>gDbP{)PxcHp`@kWGsI)HRQ7ihtPustn6>X#*7s-CTlX%s2NIxzW`Bsn
zof});qlcWg9<g`GF?}17uD4!An?GHJQCK06p#f#&+uC!gy}JFLU#(H&-e`64YKCl{
zMo0!YbAZ+nu5=~INHtXpQFk`>95}Nj60K4VaU0E%#8)L^V>72ZOuH5ax;^0Tzuq}g
zOeM?NehLhXsH5j^0-)tZ&$rCK2b^A7HN8`gv(TITK3-+hO$R-eGWePTxu<>C!T=>4
zsY=szGXtjDQ^^yeH;M*BX$n`-UL@+k8v-Cy@d4`aXOwF|vAzhv9Pj88On}|g9->zS
z!`)4sYo;7Rru=PjhLI66rQ_qE+G6Np04*rcNMW0&{5#b4B?4{&_|6P;9OAj#yzCbe
zfPw}pZ}3U=@%An;4M?8o{iV7gUtGZ{)N@fw^I~1_wd@}|54V>$z_v%YDk#&L3uQ4r
zd~$2E1yhU!UocNjKbL9Jz!zYlq|dq4LoTP!47420pYcm}k8`Hdj^Ou9zAD`W4$0%G
zp&DMRWi6rdZw2O*@Fq4jn;wTxcTS-Zy@=pvH5LcAh0dmF9?K|VbS0O|FnU<%RX&?J
z3r6Z(F5)j>n+J6<0I$|NXYdd-JtS@?ywzK@x|z>+(1Ew_lWWT?@NFlM@I)s%aFPHy
z(H;>y-1VZI+`jUw1MVoO9D8efzo8CrhY#V&Jx&#APFqD+Nq333Twj3P2wpUvBK=fn
z$t9)WC-c9TycMXDyTT6jwLfgBI%Q@d^;D|(9KFXwz0U}~ZCQgMPpDiHEB**MfjUt+
zrym`pVzu4~GS-O!-z*eMr{nbXfH~Ri+pb)wUf-O+(k8&#;aL#C8*skw|79AEK_0P0
zQ(`3M+N8S>GriA8f6u}FZqb&WFT<vxc_QLFX(sYZu#(=rWpgggMosAjvxC)$0uDis
zO=P<uygw*UDJj~WNJ%_5)la+)s|j@=bIWkDU<Bzy^W|rRz-2^2!Jn+L*&Nt4Qugj<
z|7y0VUw_)z?6R8oV-a(k<pD+yScJFn$pQQV;)?Kg=8jD>kkycj`n6F+|Hsfb)}6{z
zB;1Kc4?7ITbs{?h6#C}6QZ>IwKpTkr31R^g^6ljN$%TEyxazBRGj8FI(Sl42RMOUz
zwL|A6N<*~e@|F1Ge?vU(PTfh#xPD=02s^JnmB^uvHf=lOu;bGL>bw!PrZhx4qP1$N
zLk3%HUqUosKBa6^(*4ld-4){Pjco0S9;VzR@W8?$qEtv<A?U*g!a%^i1mp}m{ZrG<
z>L$0KZT=F8Ej?*Bhq>~(o`RkXKT$vZv`c=3oV6s(`V%*>>Sbo-WObApSQI)N^whY1
zi|R5o5>4lZ0cf3H@Tq^AvS0fjeq!Lp6G=IugS%6o&u+}h3wKvNDP&?atgFmX)^a=d
zk~%mkwB@};_%ZBX#9JPCi0yNz@A-Ku>4o4{L6nBks17ia7PdLcjy3>wZt(WQzysUz
zYu*iq+6^9eHe?oeynp&neV-dLhu+H>_9D7lkTtTQ2{&o)4f$TV<uP1dxQ!SA#r1Qr
z3Do8!$j>It_m|R2D(?RJq&Mmdq{)|rFtFZzd7t+F!3*k_R3t?Jd@bM>@~Vb<`S?i#
z!5(&1Fs|h@oAc9<5G-VVR_sr`e`p}dTt{!<GsN;twMc=f^fi}|)rNN9WH!S4CgsLK
zmphp4zJfquMP&xwg3HwD4B1bBc&2J!;zOpx8$%qPuSIw%<k7^Shw7uYCJ8K_jlBO-
z^WWup+_f9+ciFf3EeCw6{Y;a5Vz%w~lHQYk(7$P^t(>lk<d@{}5!AU@<#w}%+DkgJ
zu=4W=Ny4;&PEPe(n|!b9NuZu<<YgR9xbYwZsZ9?~yMEdO7WRXIwTS%@=6j&rz44$T
zkJbLwH<T$b%H*k_C`b;h7=B~-{!=I*z2UlcE3aU;glsBSt_&9u@8d_ftw2+62FMy`
ztGDcO2ehSr^5xADrtnr{99ATy0(LsBv$gZ7Txo>ML5~%9c(~WC4W%EjBE!|(<KcyQ
zs(fzH%&|67=L_Fp>9Li18i1*3qg=H5rt?$VVO!r)-r0-XnGxQ-BGgYq0MHv#YQVyh
z=%R3$PSZU(S8o-Dl?7B81RWsX=s!lLQnli@ZkLS$otTFW%&J)SiQ7Hd^rw>18&hY0
zCWNM9g4B@EiKkR4Y)IwnW!$&yLazKq!bXsAEXqOPR{WE3GBy<t#vaJyL0$kpJ^-ru
z1w(Q)F$j(CO#)J@F!}+kD77DMtK;0xg#?Vmt8h;bY}t*TUQlRlXU#Ef_xo|=%D$Kx
z($S_uBJvJ8(8;5YHYD6vDeQgY+=DUo2MjQuR~LU&L91)1EEY~d5J#~b#P0|V@QazM
zS^%&ximMH+!GUjPn!4!>Qp$o}f0=F?6bpm&y|<xvd`pRW!{xAh8mY8=VSG2&l;+<i
zT}<LRexz0Hz*zO)QrX30pyN|7C^!Kn(9Ps5(y;B(j1b=83VeTv{Tn9Fmqhy3V1R!7
zL<Crd8X8+9jSXJ{@&fEdz%Q;G->hz?6`OzlxLwj;3h~BpdBhT~yO0sp@#F%!;P1mt
zAd|#9>4dq3{Q3uyBnmMuM^<47Lp=g~UuiNBeTyL?Hzx}GfXQgry+2#9Bc>KN52sIL
zB518%EY+MK<9{X1R$JQ@A-_=^?|bUz$VJqoFbw2fCp!;{2#NRGL*bq#h5UXEKp1G+
zH3+Bl;7Xm+CsZIIEBH3psdASd_!Nul&8-J6?hb%JYT4vy1q6D%5R3!R@0WM1>%gXX
z0t|@PONoz88bXdVLzn2=^>^6uAWuazx+<qzwY`sJcoT$8Ifs?#5T%rDGs7K#_>=fi
zpn=oFgIP{Gp*+Df)NdE`mAU;=<J#1fq4mAqy|O(CP=4ZPLi{P*{m=PvLQ?eIS_;H1
z)b-dL-;xxBwW=dm%i2_^2NsJr3~PH)S}s`kZPrveA^+pTc%LC<wXh;!?z?(d4|%PR
z_;c|SpxK&*u9yKz4A(P5)gmfEHK-toHF80tna=fE@b~OrNay6OiT)!9TFsc}hH*Cw
zT#*CWm7&%Rhk`sWL7qln4LVS+Pq4<$<JE@Yg=13g9dG`&`9qJ^EM>rGO#mK8uC;vA
zczyD{9Cn^#IVkHL@W8^z{Bj$ja<_fp7lTS983441FXsC~Gz-u+JnTOqWtE1Wx%{j^
zlI1kI)PO!UMDstQ)}CZ(L1m6wY@$@JIef8btr!i|FNvf=1fjd;l^NmNb#SXK$a(|r
z0x}l9f!5^!8{zIYT;|7DV|3-SA%5~C6dcK1iB?d(u<6aRV^C$b9)oHJ)OK~Q6IE{(
zn-k(|ng>8-MeOpgp5!MBkin0Q%O2OCh}55+FcJA>iV#$zm5_Q6fIyMTow`>YQXHV{
z0zpMbuoAcH(=!pbQTqR_jX30g=y7n6Uky)kkJLAq*#?|pYK#8*E7Q-{3x(D@-GqpV
zzIpP`yuTaKxN5Fd^9*6Mk;^2?HC$o$>lrN=wj3$N&+=2Gpzk*ar~WBt0X^yPK8iUf
z)yRXy8JCt7eF+0Chz$Uscak6s@G!@-55w=7<u~?_44it)<UnSDo)$uX*JL40<{PA>
zj;Wjt$5i#TJcrBfgvnB`i^sgf)Nhw?r!QRgpf}uNaVB5XtJ^WnVBiG6nXgAbw%4DB
z!Qe1Z@U}IrNU>e7KU9$^1*>Ncy=&*A2R`G;pLQ1=zeoB!th_fP)$=LxEg?f6d8}3`
zksqR<n?$qrORR|A;;%J#r~z4G3cnNztzM9;1V9)7X%^=-P7;F)(~SX$fVoo@&Gbnj
zFt!DVO`p+(fG9AX?!~_LALt%Fw|mJ#N7j);{t{ObAet?EOdn2=U|^C!H%e#qlbpC!
zKz;%UwE#qUqfjzo3XotAkvrQqq=|sU32>LnKnM=y{<S}HG?DY_wuBos=vu{{xl$-H
zq^)$Ek`?k8Ckk9+KU?WoaL0VMxQv}zR9_Do*LV8>F!8J4DdyV~%<&7Kj;gtI_iY}n
zIU;?oh9&EVzblllH|Ugf_2xIv9}eV2-DFVTw`SCQJjkPMdb1#;GC~S!b#*c=gttw1
z1<l_>^#@SAgsCzGgWTL&_J;J1C9r^#^268U*55^3-agA#z@={=-(u=j+&5BSFc_9J
zS<;xE3iwmF_|}c3C0g;)ukVM6t@X5{_-GS+>&034qM-5ObubTQ(M)j_C#{F}gNz47
z%$DV4?3Q@l9yhk7Y^5-LtaKG~OJdbMH4>FC<5^P1f>Z=z(_P|G7Kp(3zx<;mPQR^W
zv8~FT>w^P?)r{#K5>k#Ba%nf<vGq$r{Q23UKhI$T-q^enm#%aNhs3Xg=jLdnAk^Jt
z>jf}l(x&v9$l1cDgjM;}5eSc~vzfOcukT?lVc8TtMxd3JhfPq4sZG{HvQXZ%?7pd3
zHLZITxiVjphI^vzuw|S%uKU6D2>obNjeoO*+fd~yGp%IvP2;oJ!NF+#+}$VMdde^?
zt2wOVK0+yxuSt2xukUk2-)B@;_S0plFnWIbHwfcN0uUG&?$I0w(1-?uBpJw4<SJGO
zLNR3zu-nXN)JBe@Brl%eS=Rl=P8B-ZS2qYAR!klShIS*n6@5|kR_8FYIP@VP$^yky
z7Jlv#=)LB7H~+b&NUN#A?#>ll*AJIB*?F9XJH7aA!a=_T)_Wj%O{Cy#8XuZlU<hj;
zG_l=mQ^P)4S;Xc8af!;$3*&)b2bJ2BGe}0x6x;ewg|pi8Pu2n9KeWfZ|Cus72%o#z
z6e^BRvaE#_=}c`HG*Hf98`Fu-81@QS<nkRm!3P)^hsSfCSb)hv+>)TqBWfL9`0A!P
zV@jjFuej&4$%PbTh|JxzFfCWbd0u44*Y<jYu$2?!FM*u3g}lA1@!ulH;Pupj7{3Ec
z|Es+ygW0GSB!3>)X0ZfCVz|Oo<=^)2`vfwsx`iLqUIi+63N+Mg0id)Vy1oIv9(!7z
zA9eY9T_h5hui*0r)kfW<N6-d|nQ~*>m$zIAU6!N9;+j_N3-lL@2BMG3fEW%B!YZxR
zHH|}bC6#2QInWQczcB)>aT%W*=eUW^Wp9nuwVKzdh;B%z1wzli{U7iu8eFWl_Yg?!
zk&ygsyxz&ZW&a^EW$4>qx^fU9^9<qH0kOkv3g7Yv{lyy&4>gawrIp9EySA;GooisT
z(5P-gYy}CVWc2*40B%NQ53OXtu?j1Elr{jlUEqQCxQ>bUa2>CJFK?Hj&0C@GMC>iw
z^XCOGOhZDkE-!E-QQ(~z&Z;{<xkSWBwaoO?!eJS>Eq&9DALa=fEicir#4lW#FV{!V
zY@D9$8Z=~Cadd`{|K_|2@w^e(y8v=1xV#xGofUu~FGTtoL7UzHHo^YJhf_F{1<$w3
zH4|J|Rb~T>Pz6d=viD*_u7Y<MJOGa1Rf+lZ>0EyvkK;xU)Ae+m-|-^slGkp;!=`WL
ziv3&~D6LW1kdI(HhbOz$8rNXr2sUpzfVd(MplICCt!`9kC<h}<H#>XWMgyLta`#1J
zSIUJ!KexMhAkpjtB15<qWZ$=<4#Y~1VdsY`tE#Cr38DUBU}mPFr~mrv7cL?aQj-uF
zNP3|R#XPw%Z~DklExKu1Tp{@5uI^d*IAYZHXjP3#CDyr*U^fW8bA;U<2m`PC)o~!D
zj*F6a$1qPB$n)?^I_3eHjQXMaOR&n*btv66s6-)fp%iUU5SX7VcQA2)H88Pgjjk~g
z#%jy)ou8`YS8AL<=cl!W<_w*nyq9}Uz4XJ&i{CEc>M8pI1q%IhHei1b1|cyuTP=gz
zM}!4STt<L$fEhT5+#}sotOwE@@Jrh{0UXC}-#aKx#qVD}KSBZ_=MS)hKzEQKzh6yT
zk<)R4{9PdbUd$V`?+g#>t!jExAV+PLL>TkQcC@~WDA$U7O^$P=b=YN?=f;wgRVlMl
z_g7gP@bHPF__yt+9QJnPbbsj1;*8Qt?QL5PD&E~&(GQ#)qM$k_aefN2GT#C<gD~{+
zIXVwZ)`0duW)nL;*h{-!>Vws*tlDT#-6I4XbF_7MNYX><rH9?6S230-@Qa0npJ0eH
z|IuaFMIQoBQOoe1Ko?gRunY{7_rsQ(I2Wx{xY&un>~H439ZdkybnX{fW_3<?DUAS+
zXcoB=0y!&X)Luwyul}7v8>hi@;_j7@LGWKe1xK9N?5nHmRO3Zq6g>;bd;(Q-^FoG&
zMd+%eAE|it(WTm-P9oDPWom0YFG4xyy5Zn5Wh~^>sCa22`CUv>Obr2`0`&A47`w@@
z^+2{sRDL9g`&&IfC_R-OV~Wv~bGJVUf>Cwm^Mg~Xyy6X9P&9tjhGAd&_Jky3?gK+1
z2;3c{Ft8~#J-+CI%yV4CJ$<%H+w_e{(Kia7>j}by>JH-Pu9gR|zn<CnV8y^u*-^Js
z5$#`~eg^o|YPCbn@kL3^^-b$EeOO@2-}^42V6*>rzjlE)Vn5mR=5&*>=moQof!X!X
zQF@$)vOP}GXK0a(13%eK@k$+{F?Dmbu6;{BbpQM4>g-`|^xzIpD1e!@JU-FDSGL|Y
z?k($0I8LvPCqfydnEOKv#3lh4F(5}Yb@mJcUN~@Ak$}8Z+@PUWAS+l`-vlm??a{P}
z?yGM58~<@Q%ykI!jR#J<>e)4sg+~UDSO82(IvtOEkMx=!-tI1RGT4BomCzKw{h{Mm
z7kfG69x1A_7q_?1JTC`qgdlnh?2)$9mlH%FWAY>y@1e8U#4a0P6^d^y_nH59D}A=l
zaD*l1c2(4vXPcI(@Ute+KPXq;{VIPmb6Q-DupN%Wsg5pnGpW9cw!4%mi!>aOHn_xn
z_v7Gt4(OF%PmHk{`Kb)i>MuR;CpmH>K+E$BrSvChOzJGx)t)3Ex*H$h!J)*BDaTa4
z9TK;bYm0BWMpKvj8xA1RCypjGuI#M~!=-#<q~_y{RAgb9+7STqAhka82l8<+%+qHk
zzyYOuzcvYWPIzMH^>F@Uo3S|`!2f;!Gf2xKU2;8deokR!$iNThzfh^`MI=Z2u5P5c
zli3l<)XlGHq`58c<v6%4+rq3!rbD1>XSArWQZCh^7Gyf}nmAbO2hszgL6A8DGI+Z|
zte4Z`@+lBslDmT_JPh-IZsZ`=WjIFZF*;8I79k2SAK{Tw{bTVCTPFguKF&N4Gfcrk
zfPSe+1^)o)Y;OULz8oY(d2}=p`a%c>-utMkA|Y;wxP`@EGwXDLzTFKEsd~uWgGXed
zXbAPRY~G;<m0b6iswHZJ-d?b9g5C>rB4*O2TK3@4z)EtVxfr)7){uGn!bS~hT8iLx
zP+W|*KM6H0Y{HBP)zR>{6~HeBDHu2aWf)vX`Nmm=SvmV_^(}9Ce#u|sAq9KQDw62H
zn191-cm*`|4R}i^Q!s4A3O8IGGl16uOJ>o4COldM?jfFPEQS!fNU;z*BJe~sM-e{~
zlFGaOs(a?-VY6s8f4V0G)(~v-fU&GR848d464(2@z1g2l0j${Gy&&5~fF-{flu5v#
zh|;RJc4*1|69U|q;qi(diSS1hSQYKNelZb~aQ_0<8{8mVphr(P5ETjMFJAd%<U%cQ
z>W0Y1X=}RW8+7oe0E+A<pE#9VnWM_YxPtXscF*?3yg?4~bFS=<DkVM#l`adsgTEd}
zqu-M(+Di1*Kcki)a+s;I)RD~BwpWhPwv!9$%N!J05-DIIeny^!O!o%~9E~2Ye+3M;
z-}~cr7ng-KmL0%u@b%(5tbE!Qp20cayzeSooITHbBLz!xPSf}|I;*O-(SFFf4ti-=
zk~<|&MBU%EwMXa)D+rS22I)M<vwUfPRJ2>N?7;R#p?M6!nsk^qdD<n~;_nM|_koB^
z5k-cYJ#5Xqm)|42IiUlh_x|Z^>Yf8OS~d_Rk)AxC0SmiFUQ&~@@rM@nYl~ajCFs4B
z?GIt7Y1zc$@3xO#d-Y3(qY>B<hw~2HVkOi33Wz^#C4B#U!D*MX?5i})I<h}_Bh#cC
zy~*E<bR}uo_I_KGnaTl~mipT=Q3*4p6<zM``oypzO(`RG^iU@pJqQ`)ac4ftW374}
zn{6RiXM9C)mE<{`JG?}G7Xk7W6+py}h~8DS0LrTtjx^2|{%A{SuF~WaG+@B1vCml}
zXAo!yW0e(`^Go}eEU@8&s4Lak!U%zj%{C&3;nveVWQ+~WAY?0#R~48Wm}cap&s?F3
zqo#<10Z>QD>1Mv=gi@rc<&1~o-j)joKbw}&%j>AUs)%Sv){7_ubI=xs6ogqtFJezJ
zuQ2i#Av39c85Gg53u)Cv-Oc6F!w4&ED!#0ushy>wVPO5nT&Kees$vko>atry8)4=&
z!+YO9-dotW1{`xv56h<rSKT3gN5{>PAP*q!G3B`B9}^2-igs1?u5$5!^L=qn<!I^T
zF0Ny7b{4c>M$PK|_`7U{?xrtY!Tc#Hr}Bx2mPr04+;WC`I7TO>w?G%$1nJnG&J5#q
zrm554<9#Mt|NQz3q4-ydoPt^wVg^CghO~l%_|N0!mzO3aZ=krP{^mBPks<zhP@QQ)
zBw=du-}29k*PV}C$M){5w)}|qiMlmRh<>X5`Zw#P-xd*FBd+kNz%qE(_+ZOGFXKQX
z-aJ6s%6GGPC6x=+y!H?cZQuculJk1Hp>)wYsMi27RDl*=h#<V@dD$`o66L<n6l;4d
z|AN{c(vVcuB!2w;XQBFSkaF6hDpB}b7prGeQ|(E2H?CzF1=^C?yT*Bp1(^7|2O0$A
zk5yJJIl&x~D-#b6s@u+uanaq%A_mRRj<$jez|qPUMluxUA|MWLYVRLz)*zK;prsj+
zt^o>9vW@-HQwuQ(Y=fMMq~3GwZW)3%)WJ9zOvE2tvVD_`3j@o^=!YjwKlC~N5zJk*
zIZ#*2Q#*F#GPaIVzTK%YSfE?k^oK5Y8d!~)ALYv(aovzTRxLgW({I3;BH6X6ulJE$
zl1O(N@F`H~m$v7(w2Qol{!x2hz+yp+QG2;(Q(XW(&G&~Msk=<U1gtLS=tR}(%G&#G
z5eW6ahHw5<5SSKvPS@W0P$ec&#?7<=t$7llSVkyuc0`*1`gJ&_Cyr{zeltN^hHd`9
z>m_gV^OwqtytfW}`iNPUb*NJu^p~AHWAZ!U+CP`^bcxumCKj~=tuEi~!JKU3s7gv$
zmNQw0sU&mN{Q3DFKK)BRDW7g8p!UT3?>m2m4H-`Trhj^A_rR+*`MajBfO@hte0{NW
zgtUHk*XCl^^A^aslhl1Dh2)ID<E3@Dw}#?ymw=zb9liQnN$CvwK{luq)1?CSJSf0~
z=kkOjJ4z>|x^%megfi2vAwEO$R+7`N)jN(%8^jF4p}lF3V4QBauoX-sX8ddxlqdNK
zlS(z=bzfYx_IRCI69L4rd$58!YV}`p0-l(!@cdVg9a%f<cOZ&mw;W`Lf+3+6gv%KI
zdoQ^A|2`FDB@+K{z`zG6{@1yI_u5TvoFOORc=%u7z<@4eBXH!4^?dDLP_{hYIfY<T
zMu7htNMOHPdb~FX>Mm`7@#9Ncl%@FI?C4!HgVzau+Gr-QW~_s=5k9vCA`4!xM_Vt{
z$2zS2WA}UTJ~=rd74kiRR_Xd)Z_I-OAFSu9f1D1Bq{08_HMnx|!p_DO=>!Dkntbk#
zyHl@R|M#l5pW~~kO~zsG`Q*qa{R75k_l;N$h&_KwXLq;L<$~L3&#Oo>x4Z0keOXCK
zWRbh0pxb6V1>S(;weAY{V`GpR%fQ02nPk^&F`XbQC-)2uEh8%{xV5#_HRauZuiBX$
ztN^vbd`QMs+Oy@K(N8xAotARPcM^E)DE6j`t<L!1Z|wzJ@2*cH1$?{02qdJWIIL&b
zY+LSK(a_L-lS#|VV`gV(V{-p*pgk*uN&Lx#Jn=(2#8g#PgKjf3GJ1M@EzbDBFJ=nA
zo1NU#hugyn?`y|Yk4+Nw65U=Bv*dt|74x%XbRw45^z=PVUESSoM*?`@27)>0#p&(^
zX@Z-%Hy0^6O)*TmBR-F$ixp}94Xkt8qHw#*j1=wzr)bO@#K*<S;(%rqe4S?%xFlQ(
ziU|EuIv$>yw)?ueItgiMo||OgiRUKAX0u-wZSUv+XKGlSIaJJkFf=p_lK*c&zAnQ3
z<+Rb?z!~X(!Hc}~QOJ$n=#S_0xjB<cgTH;+jdTZ&*^sBe{uvb|CMJf!&d#2~Y5#wh
zUkffD0=%{UJd5r@my2GJO<tFdw+A&UN=j|#(+1eBT+_N8;3e-0KP-A3Zh#=lRbZhs
z0|(&TZLs<of$@jiC?PTN@u$l{*yX<Wp2{jJTFu^kTPbc7K`7X!4|mu4*A!piR~fu*
zdt;5&1Hb#ezQT%v$4|f*hf|8Arup9VrkJOurbZ}e5T&KKuYU6ehtJ>bDEh7b?WQp4
z`AO}4I>-}FAw&vk(jSf>7l0o>LeIiN%gr4(t|0IOyeiU)+5Ol<rT>iyENA!vVJ9N1
zq!RvPB`)F1^|V>M;r~0$6#xHEv;T^A_*l7z|5xn7Z&&1hU=aMh6?VAF@xNOr_;2|D
z;VXYh&KY>R^eMjmcfsc2d;Ehg3$##p36WJ)gyDa6<<1%jUny|J-{uw-g{P-e*<oi&
zE2}<Rs>jS7mr++IL_k0Qr?vcXij9lIrKB8=dL@mDf${&eckS^|rh9y(`0QDi${s~A
z#fqem$X#2J%2JyNVG^O_R<2`Wx*#W^h{-LwEvD9(wXM<2m=KeDC99Sh6I0_l<34UP
zbAE4~&pF$E_Ut*I^Y`)JJHzunm*4aK{l4$(`F=<5gXG?I5r@VTi!^aU#z782cmpo2
z#gd1l10Eh~6bfZ9GPCW4CTP*64MtinfOAP1UqM0PCP9I*7uGIQ)Q9IWMB8vu#y+^5
z<SiPSpf$8l<bUibAj)S$W!!ul{Gmy==vlBC9<LXlP8V@>giO(MywH16g3tJPN$Akt
zKlm}~kvUQOuc7N})KC&Oeg*>)7Zi5`r+l+Cu(XQ9J2GC8Fb~c71raG}Y4(S|s+s?g
zFPHI<)U4uS{f(28f~~D>YyQp*S$U<{zzRldUDB7i+x`CnvjtoHoMfI~Nsvn9Sgvkv
z5fFQ;RaKLjOuIGM*p0B4u<B=plM5qclju?Rm`?IgXUWt+hTRi1QN)Q*-Fhq$G}&YN
z;suL5vCxt*-*WuWR}X5<&d-I51|~T%KeV<wA@)Ueb#>zt69}vZbLely)Ft+q1q{XC
zRy=(~)xiBJShh6%@`I<^&d~K!{0?11!;YRkBT@SL`fK46a?L`}y$r<UiVnNc08786
zxXz5r=d%E)%F4@U#~txSV2^VSx{hojfZa90X6s4{qty|69X|n6nouK^;B5ik0fUe*
zS>qLcam_w9etM)k!#-7&;@_fb7RG7aTvAYn;}^07Y?=&CXyXsUB1@nyabolhMNvQ$
zG0&u_n_e9q4Ho?zHxnfCgYgyKy{iTGW*1+c;<3<#lO4Bhtffm+iHV8Sj2ldvO^SY8
zcG-~<QVa)IvUui7{hh|fMiM~Zbbo1_-7YcAD*L!qp|h!_<uNhbXsCuvrMjEuw+%yM
zqe-h~Lvc8qesQ^1^>-M|7{g3GZJ1v~5ZDF?ISlyNv#Q;*eza@%?%l6JAma!Jue^E7
zmLpdb)b2_Ew1!9DR(65KlN$yz&lU!`*v{X#4DF>lptD3ZXKa?n>xWx&P4drsNd&WS
zO7>OiABs$j(9M6W;g7$o=ka(u<~)|KSYaIU>yCMk$cym247GdD9+@|gNx(L>cF_*j
zd_g2M6cThJ%-5fvY|@3CU=+U)2{XxSw-WLmRkjC>+g@H0;OoG04dT=vv`f5j@3Zfh
zl*qzdJ_awd^rkG`h>3<ZDr#y?U^Il-CG}>s^s#T-FT0NAeXjms@9dmTbm-by%?X1J
z*?@slyok-Zre?C%CT>SDLkhso<#Od%&5M1wXhHAaAfy8-b9+K*X=J7(51^qLFO(@d
zKlz1cRjB{Fx+Ip(CLy6a&n!e8I<n19Jgox1XL&^#9q`K|kDJ-qP58bBB@N4GU~bob
zedC@v!18MGY!4VF;So$l&C_~1*V1iJfYm8hLV6aG>)o$xeZr|YQSW%*0FsRjYvwMU
zAv1CyTI2BU3=VXH3W;O>J@I&jy8(z-7ezFx1yV(<bd8puPIo#wI&v#JZ0lJpR;u=h
zxWHLAOV{NCS~@3>QrYC>(_QY3YzAoXmzH}fAvz`-7>=>A=5Y)O_3FA$SrC8PH9l)Z
zoKs}q{bZz^HuvrddTy+$Z?KzOv?v)IJoV`4mWGA~#dHYnsasB)<dijA!KV*7eGY;;
z1CVVpMK!4*Z+}dp)2%l-7o^nsJoPcEIQRYlHeudhK~DDY;lluFhr+{+z(P$&N5?)v
zL~{qc0il-Q@V=B)+Pw$4JYi^umW-pJ@AlkVJ5H}|AsWW?@gpsfMAup3BI<v>a9c)J
zZl#6>n;Ure*s<nh#dW;EmtO#tEvE_kVYWYj^o#$~3o?qDMez3y_b9cCOoBAwRcGgW
zu}}+gc3z%3PDp#xTOIMXIzksRy&wdR)+3=Mh{s87O6c?gNT&dq+RR1o;KpletZYvg
zXXi~DH_BXxU7B81$+hP%@UmQ7!7r|6Q9v8MZG(U98Cq?xsHAihoapE_KVc!`II({Y
z$$t{kWA@iW$7A2Ehyq8rpkv4KNF>seQ!lkl&CT5*9>TgbtfET+i*8%`JViYwk(yhy
zh4V-aWvZv==uPoW$}21!<LGbM!e;eG_k;vVCE~8*1Tlt9$5mRweu)CJKkWJ_;RF<X
zrCZ#}U6s+cj|6Whd*XRF_PN?0e*NGekwnrLGD^D0mmZKNB{2`=HDh-nEyvA`pNMM9
zGaUx0nAhXu;iiU$E*K2V&8x<=b(qVw#S*cQ;HJFZyf^_ufkY;g7no9UNBp&a;BgZ2
z{X;NCbo+KwmSxniYm|6lhNJl%dAcWzS@!g7&*N_FLlWue>hTVOq!QruL0(%xl%ThI
zSR!6zSPf$DQbZbu`fz*=kue5?8P<sDy_1}nSXxt4GYp#N@}^Mr;x7yAL#glNOgC=)
z#x&O{Xz$)c%NT)I%=jJBBnSU1MJ9zV_2z>W8OG1o*EOr@eq%#J8Tjv&<z;5}CwOrd
z%V)%*zK8erWgKk_sI&ysnSn8cU1q(9IJ;9yBE<y?>^~J2FS{L;kk>U(G5jt?|1<M`
zWFEh(M*f6J#InI^49?*6_xpw^!(H~S`NUoRum$r%XYdYuenCbS$$h=lv*!WXcCjZ*
zbq?kjK{f^HD&2zcWwhwApzw4<=2GV2dk!be^7lJA@c>)G#$%z}R<>I*nwXfl)fRy4
zP0uh!7xxzUB^u!-Po9sDx0raQ@|H<pE>gyNsv-b#RaUOt<>RvfakYc;(tA1`#y+V3
zvC#4Xh@*pTAOaa|T1<pgEQ;vDxKUs;i;R_R=QecMNv-qELbSBBE=9Ck$mh4oow``B
zL-eYHdc3i{&<b+9q^?K`S`YCeZ_yMPlqzf`bAAXC%nH~o1j!V7Ni1G>ZD4|h<|BnJ
z^kA)m9J1CTKQ}B+2V%(&Qfe^3$v#co@_R4eJPx}Cj{%)akX6GY3q>#a{`=$?FAjm#
zYUA`!CtK7P+4-&O(WB>DwpNA1_@<z&3LknvTz1Ia-I19fjs;>=&C5JUoowFH)X@<*
z^y-0tZ#Vs7kAcUJi7?8nf`V3n-#GwSpgEkge(n_C6}m;|x3}bM^@F%YAuJCaSQw=W
z;R*`^uYl%7Olcd$c~657c}7|OHS*HrB?}A~4|#Yrm6hDS%49O}c$hwm*zw{;>O$@Q
zQ>XOmlNJ3SE2Bo)G@_bGzz%aEBqH;0E=uXj<@@))o{qDr4D1MvHULT!C1gE&hPqB9
z$|H_|)S>V55eAm3x;4lMAyywxM!*fTiA33olTE8LGc$X80FQd%U%=iE40MtPh7HTU
z`)b9D$s;JqB#~JGqG#=vBL)fL88?UY^-Tbl?9Btfs1RLWh*0#-b3Dh^022_pK=eau
zRncgnZ78gTKn%r8W>%J(vF~z}w!<|a01m}=T0on6vXaJCEIAOWHS@xDISZXJV=!by
z%BY-e-pwo4cs+&m)_G?ND8ab(<tq6Nu)S!@mMxnbKue#)-pl}vm=CSFH*emofW}}-
z_d7bGxG-~CU<z1Ts;5a<Itz@<0xBh$SnfsnTTqI+*ok+*Cf^JcbRgRe`wkXBc7w6;
z4J2Ex3#DEJYr$L~*f(-=9I#kTqSv|oseX@0r2blYH1BTeHeW~&Cr_TNINiAk1mNnn
z;g17Jh*bQ@{>OoCtE^g83q`L*<lxHM+FGXq^BXKpd@Eyl(965_y-*~rHxIu6M0yH(
z@3%b{b<6y$R#E({8p5-Xo&oK%o_T%5D~25<=#CZ7bmK-TEb2hIm7bnnF_g~Mvv2%<
zP%pxgONxsxoGGtP?(Ox05ms|AeK|LBT>9&f5J+4~xef+w>vr#G!Q3I0mZosIK4HrH
zV1<-*Es%Es2o?G*J%aN`pFVtN45!SIZ~yHBjnC-EA2jtZru8S-hTQD8Pf7mY=%<<d
hkL>k-n$uRP>`#|RIjz%1O(^);Z|}Ufe7Db~zXJ$AW0C*>

literal 0
HcmV?d00001

diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv
new file mode 100644
index 00000000000..efcfbc37287
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv
@@ -0,0 +1,90 @@
+1.0,2.0366285569366878,0.26598543709489775,1.9111378509607304,0.328
+2.0,1.863508219437754,0.3412155143759349,1.8029196566374601,0.3632
+3.0,1.7781075000737478,0.37343204670101376,1.7365597405952091,0.3826
+4.0,1.7178380710578425,0.3947648329732425,1.6806893259419955,0.4002
+5.0,1.669877784216493,0.41048020192787105,1.6390592825854502,0.4138
+6.0,1.6350585694820974,0.42428172261924546,1.6146616228664268,0.4242
+7.0,1.6090974892669985,0.4327987369120824,1.5910762546497168,0.4314
+8.0,1.587529581892735,0.44248223782615925,1.5692068211402614,0.4374
+9.0,1.5657862599669354,0.4502518904769819,1.5481026555575144,0.4452
+10.0,1.5430813878099718,0.45822149742396545,1.5275131045309027,0.4538
+11.0,1.5199649145732006,0.46613397457204586,1.506689447582152,0.4618
+12.0,1.4982848263026722,0.4736133039720792,1.488505237593026,0.4718
+13.0,1.4789747560435527,0.48032605534319434,1.4727941864822314,0.4766
+14.0,1.4621960038234303,0.4855388898121987,1.4603915792006466,0.4802
+15.0,1.4475009074972,0.4910233505068971,1.4511544780956973,0.483
+16.0,1.4349582625238917,0.49627929200598303,1.4428575600357896,0.485
+17.0,1.423895901496466,0.4997927746385241,1.4336665640797934,0.4904
+18.0,1.414059771299443,0.5039585757021772,1.4282307999663446,0.4916
+19.0,1.404377145998948,0.5074814068472661,1.4222616630433387,0.4968
+20.0,1.3953288419933125,0.5116233172677415,1.4168349642383111,0.499
+21.0,1.386648730995609,0.5150796701013794,1.4117382901756448,0.5014
+22.0,1.37796944077023,0.5181361143426957,1.4065785507454347,0.5026
+23.0,1.3696668695223018,0.5211639936845605,1.402195989771859,0.505
+24.0,1.3616434672260533,0.5245060869203922,1.3972141118943933,0.5052
+25.0,1.3542630371676898,0.5273719253780954,1.3935068999222704,0.5078
+26.0,1.3473694290558793,0.5303141100216054,1.3907051889091704,0.5102
+27.0,1.3402073426222127,0.5322232840285857,1.3892659032226624,0.508
+28.0,1.3336872003973426,0.5347987992354993,1.3865499433195474,0.506
+29.0,1.327384157679365,0.537255380588333,1.3873974009402807,0.5034
+30.0,1.3216604201787392,0.5391406639521356,1.3836836288514893,0.5054
+31.0,1.3161280598647895,0.5419685889978394,1.3832381357516508,0.507
+32.0,1.3105416127136869,0.5436824829649327,1.3800900635777025,0.5088
+33.0,1.3047664781705213,0.5450535981386072,1.3801428125459738,0.5088
+34.0,1.2996126109982393,0.5470817059996675,1.3776929353987932,0.5116
+35.0,1.293823188412723,0.5487145795246802,1.375557900607078,0.514
+36.0,1.2884030433697722,0.5512854204753199,1.3705886811509542,0.5142
+37.0,1.2831253551292954,0.552880380588333,1.3683649044462805,0.5134
+38.0,1.2783286448400695,0.553456352833638,1.3651622891751962,0.515
+39.0,1.2733306632853874,0.5543226483297324,1.3643001109138162,0.5164
+40.0,1.268891600263991,0.5557508933023101,1.3617116557796567,0.5192
+41.0,1.264266710657484,0.557293397872694,1.3587595770290883,0.5178
+42.0,1.2596929659095248,0.5590452052517866,1.3554118331969724,0.519
+43.0,1.2553917233114704,0.560554470666445,1.3546034200212493,0.5224
+44.0,1.2516899362081932,0.5619302600963936,1.353135177682238,0.5236
+45.0,1.2473519464193588,0.5630967467176333,1.3521054237005408,0.5234
+46.0,1.2433013995256341,0.5652105492770483,1.3513343518829966,0.5234
+47.0,1.239448787494256,0.5670148121987701,1.3487089014064364,0.5268
+48.0,1.2354618536988764,0.5683812531161708,1.3461108297962203,0.5266
+49.0,1.2318261814835512,0.5694667192953299,1.3446704091638717,0.5274
+50.0,1.2282065364187986,0.5719851047033405,1.344586378891869,0.5296
+51.0,1.224133646245753,0.5726992271896294,1.3432699487117177,0.5304
+52.0,1.220546875066253,0.5752082640850922,1.3426149725440661,0.5308
+53.0,1.217120690696172,0.5763555343194283,1.3426393843545892,0.5318
+54.0,1.213568203392273,0.577783779292006,1.3404433750471476,0.5328
+55.0,1.2103219298678027,0.5786931818181817,1.3395115771735926,0.5354
+56.0,1.206984211848854,0.5799261467508725,1.3372971072601654,0.5356
+57.0,1.2037463317947639,0.5816639313611434,1.3349564292855296,0.5356
+58.0,1.2002898800179351,0.5821495346518198,1.334121740633766,0.535
+59.0,1.196807843220183,0.5832921306298819,1.3337720183945212,0.537
+60.0,1.193598377912578,0.5850345894964267,1.3323687113361222,0.5366
+61.0,1.1900622506273248,0.5858058417816187,1.331289361658365,0.5352
+62.0,1.1871378053459447,0.5867438092072461,1.3306094609968828,0.5378
+63.0,1.1841863183111265,0.5882577488781785,1.3306050856582037,0.5376
+64.0,1.1808866625784924,0.5883720084759847,1.3297544647820971,0.5372
+65.0,1.1784611619616456,0.5891718256606282,1.3293163604280938,0.539
+66.0,1.1754922767281581,0.5895146044540468,1.3281268708805458,0.539
+67.0,1.1727712033092526,0.591171368622237,1.3243216554218156,0.5398
+68.0,1.16986368405737,0.5916948853249128,1.3250558910354457,0.5402
+69.0,1.1669975974070492,0.5933277588499252,1.32206820973747,0.541
+70.0,1.1640628770424661,0.5947419810536813,1.3229685907464526,0.5398
+71.0,1.1614204583284387,0.5954083222536147,1.324079172392963,0.5418
+72.0,1.1584428785720262,0.5964034194781452,1.322592790482728,0.54
+73.0,1.155897042957851,0.5969840659797241,1.321316808597242,0.5404
+74.0,1.1533538946445323,0.5984221788266578,1.3240338764937882,0.5402
+75.0,1.150596395153521,0.5990838457703174,1.3250515243843302,0.5414
+76.0,1.1479828276607542,0.5997361642014293,1.3213816765518247,0.5408
+77.0,1.1456013991649443,0.6007359356822337,1.322209337122594,0.5414
+78.0,1.1428948834559163,0.6021402900116337,1.3208159001419877,0.5416
+79.0,1.1405840977255453,0.6030258018946318,1.3190405319298553,0.5436
+80.0,1.138456276159628,0.6046540011633704,1.319016162754993,0.5428
+81.0,1.1360969314308267,0.604801499916902,1.317674437313949,0.5436
+82.0,1.1335010399508012,0.6058584011966096,1.3186716090087884,0.5436
+83.0,1.13123556274039,0.60646761259764,1.3147843189430144,0.5432
+84.0,1.1291076858663522,0.6074959489778959,1.3154507688752388,0.5422
+85.0,1.1270488055037702,0.6081815065647332,1.3113865161079785,0.5454
+86.0,1.1246272411531353,0.6084671555592488,1.31230086333935,0.543
+87.0,1.1226729512819174,0.6089195196941998,1.3106131954976976,0.5432
+88.0,1.1202892968081963,0.6099572045870034,1.3103511664799978,0.5434
+89.0,1.11780839667264,0.6109523018115339,1.3103589073105326,0.5418
+90.0,1.115877186853028,0.6116378593983712,1.3095933331067162,0.5456
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv.mtd
new file mode 100644
index 00000000000..d13a02334c8
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_cifar.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 90,
+    "cols": 5,
+    "nnz": 450,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 19:35:45 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv
new file mode 100644
index 00000000000..5fa736eb04c
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv
@@ -0,0 +1,60 @@
+1.0,0.6600831748999706,0.7919920634920635,0.45846249291763935,0.844
+2.0,0.25601354551075184,0.923,0.3060521207758516,0.909
+3.0,0.1791216355591798,0.947125,0.21327004292691254,0.933
+4.0,0.1399477948175259,0.95775,0.1736507581045767,0.942
+5.0,0.10750665179781986,0.9675,0.1505918516055093,0.952
+6.0,0.09024092394376199,0.971875,0.15272215195884234,0.949
+7.0,0.08109318636628057,0.974,0.18253549984362463,0.943
+8.0,0.07343285225967074,0.9765,0.1963133682103849,0.936
+9.0,0.06492667107394304,0.97925,0.16139850527851352,0.948
+10.0,0.05564659771922558,0.9825,0.15813024537483578,0.949
+11.0,0.04885065099874338,0.984625,0.13685280674984188,0.958
+12.0,0.04345266142523965,0.986,0.1295635143772901,0.963
+13.0,0.03789947918193832,0.988125,0.1280999416145119,0.964
+14.0,0.03440348961873195,0.98975,0.12033225446550745,0.967
+15.0,0.03113956579003195,0.9905,0.14465871496448326,0.956
+16.0,0.02782411652521333,0.992,0.13935065575895655,0.959
+17.0,0.025903924198902143,0.992875,0.13805741037482513,0.96
+18.0,0.023049034921953607,0.994,0.1390425642677865,0.958
+19.0,0.021960489895278133,0.99375,0.14567394365139114,0.953
+20.0,0.022683080115491985,0.992875,0.13101678745364734,0.962
+21.0,0.01951068056535099,0.99375,0.16387851254291194,0.953
+22.0,0.02072305049782118,0.993375,0.13199021229927832,0.963
+23.0,0.01573298327963868,0.9955,0.11595083667347869,0.966
+24.0,0.016863144556757713,0.994625,0.12464655755935368,0.964
+25.0,0.011906263883559153,0.99675,0.09555637996964318,0.975
+26.0,0.01387731248733787,0.995,0.07513476435524133,0.973
+27.0,0.014348454929750287,0.99525,0.09799803592380817,0.973
+28.0,0.014791121236119583,0.995,0.09913795348726741,0.97
+29.0,0.01745363921146775,0.995,0.10855312385534872,0.967
+30.0,0.016810133122882962,0.994625,0.09143590778251723,0.972
+31.0,0.013776919442576465,0.995125,0.09972306275201406,0.973
+32.0,0.012081841240356836,0.995875,0.1042540514767386,0.971
+33.0,0.009673850822736923,0.997,0.08166328277502062,0.976
+34.0,0.010130896530972777,0.99775,0.08359973442180495,0.975
+35.0,0.009486136169168992,0.996375,0.09465088299135849,0.969
+36.0,0.011784477117724483,0.9964980158730159,0.12253367435178857,0.964
+37.0,0.01285823612554023,0.9955,0.0778909504372984,0.973
+38.0,0.005531593403143221,0.99875,0.08527911368874061,0.978
+39.0,0.004174380412227944,0.99925,0.06256664584964217,0.98
+40.0,0.003724836264838745,0.999375,0.06596277050627802,0.977
+41.0,0.00373747439026559,0.999,0.106305429921357,0.975
+42.0,0.00446614766949216,0.99875,0.06967587883698868,0.983
+43.0,0.0033526953383332878,0.999125,0.06330396928658376,0.984
+44.0,0.0028230799970961214,0.999375,0.08233284112176356,0.979
+45.0,0.0027498272954112166,0.99975,0.06967829332529306,0.983
+46.0,0.0021902541798457316,0.9995,0.12228901254433457,0.971
+47.0,0.014584827217146804,0.994625,0.11751439804792264,0.973
+48.0,0.003940440886971421,0.999125,0.07298610902463333,0.982
+49.0,0.002838167116175086,0.999375,0.05795638638785478,0.986
+50.0,0.0024289928108280536,0.99925,0.07105060997549256,0.98
+51.0,8.843184635752194E-4,1.0,0.06353335404859009,0.984
+52.0,6.38977018076913E-4,1.0,0.06250178394164058,0.986
+53.0,0.0010020022685720159,0.99975,0.08592793990596928,0.977
+54.0,0.003981127656437652,0.99875,0.14202391884937093,0.964
+55.0,0.009452855843030253,0.99725,0.11272917235094357,0.969
+56.0,0.02030383425854622,0.994,0.22102710131762598,0.961
+57.0,0.0201962723872667,0.992625,0.0746092794797957,0.98
+58.0,0.012778119867117035,0.99625,0.08488781163569933,0.98
+59.0,0.005098323074058392,0.998625,0.0705106763422874,0.983
+60.0,0.002205910690743962,0.9995,0.06720935609152047,0.985
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv.mtd
new file mode 100644
index 00000000000..04598b34d75
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_adam_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-18 12:40:31 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv
new file mode 100644
index 00000000000..06ae4237297
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv
@@ -0,0 +1,90 @@
+1.0,2.308857970725247,0.09733879009473159,2.303781414916082,0.1044
+2.0,2.304008930124782,0.09928120325743726,2.3033522297439544,0.1044
+3.0,2.3037258759929133,0.09828143177663287,2.303188402061223,0.1044
+4.0,2.3035719744954597,0.09843931776632873,2.3031009169012147,0.1044
+5.0,2.3034703247407293,0.09878209655974737,2.303046278570225,0.1044
+6.0,2.303396340692825,0.09849644756523183,2.303008793363021,0.1044
+7.0,2.303339197122346,0.09803940917400697,2.3029813853473597,0.1044
+8.0,2.303293240514144,0.09798227937510387,2.3029603952712767,0.1044
+9.0,2.3032551783927606,0.097525240983879,2.302943743402374,0.1044
+10.0,2.303222941068266,0.09741098138607279,2.3029301622777862,0.1044
+11.0,2.30319515139901,0.0971538972910088,2.3029188364644875,0.1044
+12.0,2.3031708517290452,0.09735385158716968,2.3029092180000568,0.1044
+13.0,2.303149352177799,0.09749667608442746,2.30290092519086,0.1044
+14.0,2.303130141169306,0.09738241648662123,2.3028936839986205,0.1044
+15.0,2.303112830057921,0.09746811118497589,2.3028872925240407,0.1044
+16.0,2.303097117460817,0.0978394548778461,2.3028815986504196,0.1044
+17.0,2.303082765498494,0.09755380588333055,2.3028764855041284,0.1044
+18.0,2.303069583506499,0.09738241648662123,2.302871861726882,0.1044
+19.0,2.303057416587966,0.09738241648662123,2.3028676548032334,0.1044
+20.0,2.3030461373916387,0.09772519528003988,2.3028638063815388,0.1044
+21.0,2.303035640092217,0.09789658467674921,2.302860268927418,0.1044
+22.0,2.303025835907424,0.09781088997839454,2.3028570032873428,0.1044
+23.0,2.3030166497080473,0.09775376017949143,2.302853976886092,0.1044
+24.0,2.303008017419134,0.09758237078278212,2.302851162373504,0.1044
+25.0,2.3029998840026313,0.09763950058168522,2.3028485365948654,0.1044
+26.0,2.3029922018735722,0.09755380588333055,2.3028460797978676,0.0958
+27.0,2.3029849296437037,0.09726815688881502,2.3028437750148405,0.0958
+28.0,2.302978031115318,0.09718246219046035,2.3028416075764797,0.0958
+29.0,2.3029714744683667,0.09718246219046035,2.302839564725359,0.0958
+30.0,2.302965231598419,0.09675398869868705,2.3028376353059916,0.0958
+31.0,2.3029592775733714,0.09721102708991192,2.3028358095141903,0.0958
+32.0,2.3029535901845293,0.09703963769320259,2.3028340786928005,0.0958
+33.0,2.3029481495732043,0.09718246219046035,2.3028324351640124,0.0958
+34.0,2.302942937918265,0.09683968339704171,2.3028308720907593,0.0958
+35.0,2.3029379391731855,0.09683968339704171,2.3028293833614257,0.0958
+36.0,2.302933138843571,0.09695394299484793,2.3028279634933666,0.0958
+37.0,2.30292852379795,0.09689681319594481,2.302826607551716,0.0958
+38.0,2.3029240821061667,0.09692537809539638,2.3028253110806913,0.0958
+39.0,2.3029198029006626,0.09689681319594481,2.302824070045189,0.0958
+40.0,2.302915676256911,0.09695394299484793,2.3028228807808824,0.0958
+41.0,2.302911693089989,0.09701107279375104,2.3028217399514026,0.0958
+42.0,2.3029078450647407,0.09695394299484793,2.302820644511436,0.0958
+43.0,2.302904124517459,0.09695394299484793,2.302819591674796,0.0958
+44.0,2.302900524387369,0.09695394299484793,2.3028185788866904,0.0958
+45.0,2.30289703815653,0.09686824829649326,2.3028176037995474,0.0958
+46.0,2.302893659796928,0.09669685889978394,2.3028166642518713,0.0958
+47.0,2.3028903837237644,0.09695394299484793,2.302815758249686,0.0958
+48.0,2.3028872047540805,0.09695394299484793,2.302814883950203,0.0958
+49.0,2.302884118070067,0.09701107279375104,2.3028140396473953,0.0958
+50.0,2.302881119186381,0.09709676749210569,2.3028132237592303,0.0958
+51.0,2.302878203920975,0.09741098138607279,2.302812434816333,0.0958
+52.0,2.302875368369051,0.09726815688881502,2.302811671451892,0.0958
+53.0,2.3028726088796594,0.09712533239155725,2.3028109323926556,0.0958
+54.0,2.302869922034726,0.09695394299484793,2.302810216450874,0.0958
+55.0,2.3028673046301313,0.09701107279375104,2.3028095225170846,0.0958
+56.0,2.302864753658669,0.09692537809539638,2.3028088495536134,0.0958
+57.0,2.302862266294632,0.0971538972910088,2.3028081965887446,0.0958
+58.0,2.3028598398798685,0.0971538972910088,2.302807562711441,0.0958
+59.0,2.30285747191109,0.09726815688881502,2.3028069470665855,0.0958
+60.0,2.3028551600284204,0.09726815688881502,2.302806348850665,0.0958
+61.0,2.3028529020048816,0.09726815688881502,2.302805767307857,0.0958
+62.0,2.3028506957368564,0.09741098138607279,2.3028052017264735,0.0958
+63.0,2.3028485392353937,0.09732528668771813,2.3028046514357197,0.0958
+64.0,2.3028464306182124,0.09743954628552434,2.3028041158027452,0.0958
+65.0,2.3028443681023987,0.09769663038058833,2.3028035942299465,0.0958
+66.0,2.302842349997736,0.09772519528003988,2.3028030861525015,0.0958
+67.0,2.3028403747005197,0.09766806548113678,2.30280259103611,0.0958
+68.0,2.3028384406879288,0.09761093568223367,2.3028021083749253,0.0958
+69.0,2.3028365465127987,0.09755380588333055,2.3028016376896474,0.0958
+70.0,2.302834690798834,0.09758237078278212,2.3028011785257787,0.0958
+71.0,2.302832872236157,0.09749667608442746,2.302800730452008,0.0958
+72.0,2.3028310895772277,0.09772519528003988,2.302800293058731,0.0958
+73.0,2.3028293416330525,0.09769663038058833,2.3027998659566746,0.0958
+74.0,2.3028276272696626,0.097782325078943,2.302799448775633,0.0958
+75.0,2.302825945404852,0.09786801977729766,2.3027990411632944,0.0958
+76.0,2.302824295005185,0.09801084427455542,2.302798642784158,0.0958
+77.0,2.3028226750831555,0.09841075286687717,2.3027982533185245,0.0958
+78.0,2.3028210846945876,0.09838218796742562,2.3027978724615648,0.0958
+79.0,2.3028195229362116,0.09826792836961941,2.3027974999224523,0.0958
+80.0,2.302817988943382,0.09806797407345853,2.302797135423554,0.0958
+81.0,2.3028164818879793,0.09818223367126475,2.302796778699684,0.0958
+82.0,2.3028150009764223,0.09830064816353665,2.302796429497398,0.0958
+83.0,2.302813545447827,0.09830064816353665,2.3027960875743467,0.0958
+84.0,2.302812114572304,0.09818638856573043,2.3027957526986635,0.0958
+85.0,2.302810707649296,0.09815782366627888,2.3027954246483984,0.0958
+86.0,2.3028093240061196,0.09818638856573043,2.302795103210983,0.0958
+87.0,2.302807962996513,0.09804356406847267,2.3027947881827373,0.0958
+88.0,2.3028066239993126,0.09804356406847267,2.3027944793684005,0.0958
+89.0,2.3028053064172282,0.09784360977231178,2.3027941765806963,0.0958
+90.0,2.3028040096756257,0.09795786937011801,2.3027938796399248,0.0958
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv.mtd
new file mode 100644
index 00000000000..85c0deb5c72
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_cifar.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 90,
+    "cols": 5,
+    "nnz": 450,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 14:31:06 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv
new file mode 100644
index 00000000000..ded8beac83a
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv
@@ -0,0 +1,60 @@
+1.0,1.529923114957812,0.615625,0.4163508024364405,0.873
+2.0,0.25060122984207545,0.927,0.22952393475133331,0.916
+3.0,0.1548000660061219,0.95425,0.16200895599016565,0.94
+4.0,0.11335569472656531,0.964625,0.13560831935242834,0.955
+5.0,0.09042976548413022,0.971625,0.12328820920087886,0.963
+6.0,0.07636142457008835,0.976625,0.11456082728594813,0.968
+7.0,0.06691350664721099,0.979625,0.109108661140788,0.969
+8.0,0.05926908260776385,0.9825,0.10503831758586746,0.971
+9.0,0.053242110795777134,0.98375,0.1017104393824572,0.971
+10.0,0.048538005602347,0.985125,0.09997651189079403,0.971
+11.0,0.04440544159683867,0.98575,0.09838001759437191,0.971
+12.0,0.040599887346924705,0.98725,0.09778095571436933,0.972
+13.0,0.03734205810908477,0.98825,0.09713194222471225,0.974
+14.0,0.03431520433747054,0.98875,0.09692151531360448,0.974
+15.0,0.0316160967130946,0.98975,0.09732424066243325,0.976
+16.0,0.029327893562393736,0.990125,0.09678624020161315,0.974
+17.0,0.02731709621117975,0.990875,0.09601975312839364,0.975
+18.0,0.025400220810051326,0.99125,0.09493154669329465,0.975
+19.0,0.023849761916733277,0.992375,0.09431458155105914,0.975
+20.0,0.022359802493171675,0.992875,0.09444793298522774,0.975
+21.0,0.02104063762906861,0.993375,0.0940675280289127,0.976
+22.0,0.019815579920937687,0.99375,0.09397031760585552,0.976
+23.0,0.018703144105318287,0.994375,0.09368188701034882,0.976
+24.0,0.01755765508691582,0.995,0.09316670018928445,0.976
+25.0,0.016504243053257428,0.995375,0.09312778676414861,0.976
+26.0,0.015547213662833325,0.996,0.09313668530584004,0.976
+27.0,0.01464135909303369,0.99625,0.09262190180664727,0.975
+28.0,0.013849651145544097,0.996375,0.09243578056237582,0.975
+29.0,0.013045399463127353,0.99675,0.09173679387095408,0.974
+30.0,0.012308034427099244,0.996875,0.09119741075842759,0.974
+31.0,0.01160452532703958,0.997375,0.09061579657563777,0.973
+32.0,0.010950352937654438,0.997625,0.08979792576235757,0.973
+33.0,0.010404225456806958,0.997625,0.08958915530484804,0.973
+34.0,0.00993531874403182,0.99775,0.08940260293029532,0.973
+35.0,0.009447877428045336,0.99775,0.08878696599496204,0.973
+36.0,0.008965917585343961,0.997875,0.08823171309031336,0.974
+37.0,0.008521583132989957,0.99825,0.08802536709390377,0.973
+38.0,0.008150154841642552,0.998375,0.08802838909372968,0.973
+39.0,0.007784229895554334,0.9985,0.08738821101087546,0.973
+40.0,0.00743699102601898,0.9985,0.08739398647224264,0.973
+41.0,0.007119215903327162,0.998625,0.0871615661765243,0.973
+42.0,0.006813014561591904,0.998625,0.08662144756724738,0.974
+43.0,0.006518437802804398,0.99875,0.08661521760369693,0.975
+44.0,0.006218632411764271,0.999,0.08645243665206297,0.973
+45.0,0.005935930756546657,0.99925,0.08618233389590946,0.973
+46.0,0.005689173929135219,0.99925,0.08599448297307472,0.973
+47.0,0.0054517521228319875,0.99925,0.08606674894466114,0.973
+48.0,0.005212159802743296,0.99925,0.08593790211989807,0.973
+49.0,0.0049843212661704744,0.99925,0.08585662017134629,0.973
+50.0,0.00477715303486137,0.999375,0.08578410203580478,0.973
+51.0,0.0045963000060570336,0.999375,0.08605111424505173,0.973
+52.0,0.004398328314693466,0.999375,0.0858784041018631,0.974
+53.0,0.004202551847212241,0.9995,0.08593016807062147,0.974
+54.0,0.004025727697865245,0.9995,0.08563104128541399,0.974
+55.0,0.003846586555499973,0.999625,0.08534700218134265,0.975
+56.0,0.0036902008756122787,0.99975,0.08535591916811447,0.975
+57.0,0.0035283367720043307,0.999875,0.08520610388234004,0.975
+58.0,0.0033746694849511975,1.0,0.0851791314692245,0.976
+59.0,0.0032474977805542946,1.0,0.08501889027122635,0.976
+60.0,0.00312400676742454,1.0,0.08481266555259098,0.977
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv.mtd
new file mode 100644
index 00000000000..e25dd89e1b6
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_diagonal_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 12:04:45 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv
new file mode 100644
index 00000000000..d0762291631
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv
@@ -0,0 +1,90 @@
+1.0,2.303354277121383,0.09927237410669769,2.3026645264806,0.088
+2.0,2.3024429019304438,0.09835829732424797,2.302395347533886,0.0912
+3.0,2.3022537470495514,0.09710144174837959,2.3022451434900346,0.092
+4.0,2.3020886355946986,0.10289492271896293,2.302096100276622,0.1392
+5.0,2.3019041482290583,0.12017668688715306,2.3019092225507234,0.158
+6.0,2.301688531255859,0.13920090992188797,2.301681104703284,0.145
+7.0,2.3014280018530657,0.14300004154894466,2.301399021947866,0.1332
+8.0,2.301109485417331,0.14231448396210736,2.301052050694326,0.1334
+9.0,2.3007171148037786,0.14319999584510554,2.300619124932079,0.1382
+10.0,2.3002324038562887,0.1473133413661293,2.3000753268846386,0.1498
+11.0,2.2996250406346364,0.15494951803224197,2.299390727424242,0.1668
+12.0,2.298875319802741,0.16842747631710153,2.298581855353828,0.1826
+13.0,2.2979581473242936,0.18157667857736415,2.2975821039931352,0.1988
+14.0,2.296838431406206,0.19601130131294664,2.2963575161816574,0.2086
+15.0,2.295479784689714,0.2057093443576533,2.2948736290852683,0.2054
+16.0,2.2938451276224763,0.21202218713644672,2.2931041844251396,0.2144
+17.0,2.2919033381267577,0.21262205002492934,2.291004362607398,0.215
+18.0,2.2896238508484257,0.21265061492438092,2.288558037422181,0.2122
+19.0,2.286979065939002,0.21156514874522186,2.285732886786278,0.2112
+20.0,2.2839381179310227,0.21030829316935348,2.2824818251091212,0.2096
+21.0,2.2804795042279915,0.20924204337709823,2.2788294933816533,0.2084
+22.0,2.27661619286966,0.2084422261924547,2.27476184874041,0.2082
+23.0,2.27233921479588,0.20778990776134285,2.270293209378558,0.2086
+24.0,2.267662648526721,0.20636166278876517,2.2654318172890657,0.2086
+25.0,2.2626192756606813,0.207332869370118,2.260219881002514,0.2076
+26.0,2.2572590182860295,0.20756138856573042,2.2547077733439878,0.2098
+27.0,2.2516373187233776,0.2081326865547615,2.2489572231627677,0.2108
+28.0,2.245814456526489,0.20861828984543793,2.2430298475130375,0.212
+29.0,2.239854898346882,0.20981801562240318,2.23699057143831,0.215
+30.0,2.2338232484515137,0.21070352750540136,2.2308996938500947,0.216
+31.0,2.227779156047962,0.21227459697523682,2.2248143943377916,0.2164
+32.0,2.221777624187474,0.21447409423300648,2.2187839689159294,0.2168
+33.0,2.215869242739315,0.21642118165198604,2.212859492208603,0.219
+34.0,2.2100970410566547,0.21776373192620907,2.207076999483368,0.221
+35.0,2.2044948909552398,0.2199632291839787,2.201472196746826,0.222
+36.0,2.199092194103585,0.2212818888150241,2.1960714225106948,0.2244
+37.0,2.1939100113217833,0.22248161459198937,2.190890716689774,0.2252
+38.0,2.18896049922426,0.22430976815688883,2.185943191743232,0.2272
+39.0,2.184248558275656,0.22576657802891809,2.181228442466861,0.2278
+40.0,2.1797760513002586,0.22645213561575536,2.1767469219366973,0.2296
+41.0,2.1755397102667793,0.22762329649326907,2.1724977070890747,0.23
+42.0,2.1715336259991895,0.22870876267242812,2.1684688117747153,0.2324
+43.0,2.167749590700271,0.22910867126474987,2.164655458022773,0.233
+44.0,2.16417730105029,0.22999418314774805,2.1610486141209355,0.2332
+45.0,2.1608063211125814,0.23087969503074623,2.157636642520351,0.2326
+46.0,2.1576246790351115,0.2315652526175835,2.154407637714963,0.2342
+47.0,2.154617104468484,0.23242219960113014,2.1513416793938696,0.235
+48.0,2.1517599169780426,0.23252711068638857,2.1484218399487305,0.2358
+49.0,2.149066816222913,0.23321266827322584,2.145675349150036,0.2364
+50.0,2.1465245866633307,0.23366970666445072,2.1430793565890967,0.2378
+51.0,2.14412072766194,0.2344409589496427,2.140617606284286,0.239
+52.0,2.1418439754982397,0.2351836463353831,2.138280235881615,0.2398
+53.0,2.139685512113253,0.2356121198271564,2.13605717653691,0.2402
+54.0,2.137635492810324,0.23606915821838126,2.133940270945409,0.2408
+55.0,2.1356854486646553,0.23681184560412166,2.13192250053105,0.2404
+56.0,2.1338272048501303,0.23744027339205587,2.1299960314510455,0.242
+57.0,2.1320540810077424,0.23829722037560247,2.1281568242149995,0.2426
+58.0,2.1303613659253196,0.23858286937011802,2.126397779975742,0.2434
+59.0,2.1287450851810075,0.23869712896792422,2.124716807678159,0.2438
+60.0,2.1271998321934493,0.23912560245969755,2.1231064529381376,0.2442
+61.0,2.1257209692206427,0.23955407595147085,2.1215637509056395,0.2446
+62.0,2.1243035036437874,0.24003967924214728,2.1200814646668116,0.2452
+63.0,2.122943037851745,0.24066810703008146,2.118657105482082,0.2452
+64.0,2.121635884560944,0.24098232092404856,2.1172868604212627,0.2456
+65.0,2.120378615583713,0.2413822295163703,2.1159680410580592,0.2456
+66.0,2.1191679446234666,0.24166787851088584,2.1146974962627203,0.2464
+67.0,2.1180009668346935,0.24209635200265914,2.113471173048431,0.2468
+68.0,2.116874691196301,0.24218672095728766,2.1122863875401126,0.2474
+69.0,2.1157867655260487,0.24244380505235166,2.111141636758149,0.248
+70.0,2.1147352368261987,0.24270088914741564,2.1100332378539677,0.2482
+71.0,2.113718002806775,0.24307690709655977,2.1089602225780317,0.249
+72.0,2.112733184646633,0.2436767699850424,2.107919882704548,0.2494
+73.0,2.1117789394143482,0.24381959448230017,2.106913026103143,0.2488
+74.0,2.110853316989593,0.24421950307462192,2.1059350834636765,0.2498
+75.0,2.109955079089359,0.2445290427123151,2.104985952301633,0.2502
+76.0,2.109082620696774,0.24475756190792752,2.1040617864167523,0.251
+77.0,2.1082346435077755,0.24478612680737907,2.1031627040313423,0.2512
+78.0,2.1074095248882703,0.2450146460029915,2.1022870982946626,0.252
+79.0,2.1066062067836695,0.24510034070134618,2.1014329767324553,0.2518
+80.0,2.1058225393256165,0.24535742479641015,2.100595247228801,0.252
+81.0,2.1050574606919588,0.24572876848928035,2.099777698683264,0.2526
+82.0,2.1043111830414234,0.24621437177995678,2.0989845437353187,0.2532
+83.0,2.1035842211982745,0.24630006647831143,2.0982143244746703,0.254
+84.0,2.102876376387279,0.24672853997008476,2.0974656861364447,0.253
+85.0,2.10218567299019,0.24707131876350338,2.0967362386813306,0.2536
+86.0,2.1015113351260197,0.24721414326076116,2.096022887408198,0.2534
+87.0,2.100852094441317,0.24738553265747049,2.095325468476296,0.2536
+88.0,2.1002072680052266,0.24758548695363136,2.094641673375884,0.2538
+89.0,2.0995766034511516,0.2474997922552767,2.093973751766461,0.2544
+90.0,2.098959302207614,0.24758548695363136,2.0933196838290935,0.255
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv.mtd
new file mode 100644
index 00000000000..7bf18c4774f
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_cifar.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 90,
+    "cols": 5,
+    "nnz": 450,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 16:41:31 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv
new file mode 100644
index 00000000000..61c471f700d
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv
@@ -0,0 +1,60 @@
+1.0,1.866423388179165,0.5742242063492063,1.6439261632170747,0.64
+2.0,1.4928910617188813,0.7032321428571429,1.3853969008124252,0.683
+3.0,1.272710742254421,0.7394861111111111,1.2131369327400596,0.711
+4.0,1.1254325674232708,0.7601150793650794,1.095144211752259,0.727
+5.0,1.022459770574194,0.7726170634920635,1.0101238395751708,0.74
+6.0,0.946530560718444,0.7826170634920635,0.9456451197422426,0.751
+7.0,0.8880268839828795,0.7903670634920635,0.8949212750028692,0.765
+8.0,0.8415278206758608,0.7978670634920635,0.854089316705041,0.768
+9.0,0.8035827965978666,0.8036190476190476,0.8202159641777057,0.774
+10.0,0.7718674883077749,0.8092440476190476,0.7916438039580443,0.781
+11.0,0.7448529883790529,0.8146190476190476,0.7670526565955637,0.792
+12.0,0.721458277224362,0.8191190476190476,0.745724482535323,0.798
+13.0,0.7009611157389515,0.8232440476190476,0.7268900229389159,0.801
+14.0,0.6827906144486946,0.8262440476190476,0.7101796206958488,0.803
+15.0,0.6665491364904129,0.8307460317460317,0.6951588756175966,0.804
+16.0,0.6518900515924803,0.8338710317460317,0.6816308662811724,0.807
+17.0,0.6385824868896167,0.8378710317460317,0.6692585415240807,0.809
+18.0,0.626401446142148,0.8412460317460317,0.657986591683656,0.81
+19.0,0.6152151008879067,0.8434960317460317,0.6475896323415835,0.81
+20.0,0.6048551240028875,0.8463710317460317,0.6380074275770728,0.816
+21.0,0.595226790651475,0.8477460317460317,0.6291030696035893,0.819
+22.0,0.5862262284167258,0.8484960317460317,0.6207762320382855,0.821
+23.0,0.5777942203740585,0.8506210317460317,0.6129073353307255,0.824
+24.0,0.569829336052029,0.8521210317460317,0.6055186920462967,0.827
+25.0,0.562322087467323,0.8549960317460317,0.5984862630477912,0.829
+26.0,0.5551972532193233,0.8561210317460317,0.5918507350880778,0.831
+27.0,0.5483922690557319,0.8568710317460317,0.5854528969297884,0.83
+28.0,0.5417717020101807,0.8577460317460317,0.5791749719268132,0.834
+29.0,0.5353834920310123,0.8588710317460317,0.573021548269845,0.838
+30.0,0.5291950049277978,0.8594980158730159,0.5671532053783704,0.84
+31.0,0.5232429187125389,0.8608730158730159,0.561629254937016,0.841
+32.0,0.5176039971163227,0.8618730158730159,0.5564097976587178,0.843
+33.0,0.5122833004991232,0.8626230158730159,0.5514111670120752,0.844
+34.0,0.5072275346885798,0.8628730158730159,0.5466137422719917,0.846
+35.0,0.5023701417154098,0.864998015873016,0.5420622512278893,0.846
+36.0,0.49771279900284227,0.8653730158730158,0.5377322462760586,0.848
+37.0,0.49325002706494225,0.8657480158730159,0.5335120751616993,0.847
+38.0,0.4889316669931411,0.8663730158730159,0.5294598410779857,0.847
+39.0,0.4847795598700526,0.8673730158730159,0.5256187227433868,0.847
+40.0,0.4808159819962951,0.8683730158730159,0.5218443641381453,0.848
+41.0,0.4769954353746303,0.8692480158730159,0.5181261701052222,0.847
+42.0,0.4732806998370051,0.8706230158730159,0.514536341695689,0.847
+43.0,0.4696791877007013,0.8717480158730159,0.5111297293916626,0.847
+44.0,0.466207952568916,0.8723730158730159,0.5078820073977189,0.847
+45.0,0.4628588261989359,0.8731230158730159,0.504615170638606,0.848
+46.0,0.45960234279588047,0.8742480158730159,0.5016026210614756,0.848
+47.0,0.4564797005612003,0.8744980158730159,0.49853096066258135,0.848
+48.0,0.45340866555174103,0.8748730158730159,0.4956449103544591,0.848
+49.0,0.4504369793890093,0.8758730158730159,0.49280215847599296,0.848
+50.0,0.4475359327883268,0.8763730158730159,0.490056720180492,0.849
+51.0,0.4447109415788691,0.8772480158730159,0.48734386450053985,0.85
+52.0,0.44194446747002286,0.8773730158730159,0.48479191333510535,0.853
+53.0,0.439268379013645,0.8781230158730159,0.4822259811676885,0.855
+54.0,0.4366447818848977,0.8789980158730158,0.47979430296590536,0.855
+55.0,0.4341039553458775,0.879875,0.47737496308325084,0.856
+56.0,0.4316169860382033,0.8805,0.47503514269851493,0.857
+57.0,0.4291959500256212,0.881375,0.4727302757008901,0.857
+58.0,0.42682299095231074,0.88225,0.4704963291546497,0.858
+59.0,0.4245037978579831,0.882875,0.46832729836663517,0.858
+60.0,0.4222409407797942,0.883,0.46619377517871974,0.859
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv.mtd
new file mode 100644
index 00000000000..bc5790ab0e1
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_diagonal_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 12:53:09 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv
new file mode 100644
index 00000000000..74604775d39
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv
@@ -0,0 +1,60 @@
+1.0,1.866423388179165,0.5742242063492063,1.6439261632170745,0.64
+2.0,1.492891061718881,0.7032321428571429,1.385396900812425,0.683
+3.0,1.2727107422544208,0.7394861111111111,1.2131369327400594,0.711
+4.0,1.1254325674232704,0.7601150793650794,1.0951442117522587,0.727
+5.0,1.0224597705741936,0.7726170634920635,1.0101238395751704,0.74
+6.0,0.9465305607184437,0.7826170634920635,0.9456451197422422,0.751
+7.0,0.8880268839828789,0.7903670634920635,0.8949212750028689,0.765
+8.0,0.84152782067586,0.7978670634920635,0.8540893167050403,0.768
+9.0,0.803582796597866,0.8036190476190476,0.8202159641777054,0.774
+10.0,0.7718674883077745,0.8092440476190476,0.7916438039580438,0.781
+11.0,0.7448529883790521,0.8146190476190476,0.7670526565955632,0.792
+12.0,0.7214582772243614,0.8191190476190476,0.7457244825353226,0.798
+13.0,0.7009611157389508,0.8232440476190476,0.7268900229389155,0.801
+14.0,0.6827906144486944,0.8262440476190476,0.7101796206958487,0.803
+15.0,0.6665491364904125,0.8307460317460317,0.6951588756175962,0.804
+16.0,0.6518900515924798,0.8338710317460317,0.6816308662811721,0.807
+17.0,0.6385824868896166,0.8378710317460317,0.6692585415240803,0.809
+18.0,0.6264014461421475,0.8412460317460317,0.6579865916836556,0.81
+19.0,0.6152151008879062,0.8434960317460317,0.6475896323415832,0.81
+20.0,0.604855124002887,0.8463710317460317,0.6380074275770723,0.816
+21.0,0.5952267906514745,0.8477460317460317,0.629103069603589,0.819
+22.0,0.5862262284167256,0.8484960317460317,0.6207762320382852,0.821
+23.0,0.5777942203740583,0.8506210317460317,0.6129073353307254,0.824
+24.0,0.5698293360520286,0.8521210317460317,0.6055186920462965,0.827
+25.0,0.5623220874673228,0.8549960317460317,0.5984862630477912,0.829
+26.0,0.5551972532193232,0.8561210317460317,0.5918507350880776,0.831
+27.0,0.5483922690557316,0.8568710317460317,0.5854528969297881,0.83
+28.0,0.5417717020101804,0.8577460317460317,0.5791749719268128,0.834
+29.0,0.5353834920310118,0.8588710317460317,0.5730215482698446,0.838
+30.0,0.5291950049277971,0.8594980158730159,0.5671532053783702,0.84
+31.0,0.5232429187125387,0.8608730158730159,0.5616292549370158,0.841
+32.0,0.5176039971163221,0.8618730158730159,0.5564097976587177,0.843
+33.0,0.5122833004991226,0.8626230158730159,0.5514111670120748,0.844
+34.0,0.5072275346885796,0.8628730158730159,0.5466137422719916,0.846
+35.0,0.5023701417154097,0.864998015873016,0.5420622512278892,0.846
+36.0,0.497712799002842,0.8653730158730158,0.5377322462760586,0.848
+37.0,0.4932500270649419,0.8657480158730159,0.5335120751616992,0.847
+38.0,0.4889316669931409,0.8663730158730159,0.5294598410779855,0.847
+39.0,0.4847795598700526,0.8673730158730159,0.525618722743387,0.847
+40.0,0.48081598199629483,0.8683730158730159,0.5218443641381452,0.848
+41.0,0.4769954353746303,0.8692480158730159,0.5181261701052222,0.847
+42.0,0.473280699837005,0.8706230158730159,0.514536341695689,0.847
+43.0,0.4696791877007012,0.8717480158730159,0.5111297293916625,0.847
+44.0,0.46620795256891595,0.8723730158730159,0.5078820073977189,0.847
+45.0,0.46285882619893565,0.8731230158730159,0.504615170638606,0.848
+46.0,0.4596023427958802,0.8742480158730159,0.5016026210614756,0.848
+47.0,0.45647970056120024,0.8744980158730159,0.4985309606625814,0.848
+48.0,0.45340866555174086,0.8748730158730159,0.4956449103544591,0.848
+49.0,0.4504369793890092,0.8758730158730159,0.49280215847599296,0.848
+50.0,0.44753593278832665,0.8763730158730159,0.49005672018049207,0.849
+51.0,0.4447109415788691,0.8772480158730159,0.48734386450053974,0.85
+52.0,0.44194446747002264,0.8773730158730159,0.48479191333510524,0.853
+53.0,0.43926837901364496,0.8781230158730159,0.48222598116768844,0.855
+54.0,0.43664478188489775,0.8789980158730158,0.4797943029659054,0.855
+55.0,0.43410395534587737,0.879875,0.4773749630832509,0.856
+56.0,0.43161698603820314,0.8805,0.4750351426985147,0.857
+57.0,0.4291959500256211,0.881375,0.4727302757008901,0.857
+58.0,0.42682299095231074,0.88225,0.4704963291546497,0.858
+59.0,0.42450379785798303,0.882875,0.4683272983666352,0.858
+60.0,0.4222409407797941,0.883,0.4661937751787198,0.859
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv.mtd
new file mode 100644
index 00000000000..39ca9b83ca4
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_heuristic_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-18 12:30:06 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv
new file mode 100644
index 00000000000..bf5fe82079c
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv
@@ -0,0 +1,60 @@
+1.0,1.5299231149614017,0.615625,0.41635080243384964,0.873
+2.0,0.2506012298428366,0.927,0.22952393475157162,0.916
+3.0,0.15480006600605273,0.95425,0.1620089559895203,0.94
+4.0,0.11335569472662702,0.964625,0.13560831935160117,0.955
+5.0,0.09042976548384442,0.971625,0.12328820920011867,0.963
+6.0,0.07636142456988171,0.976625,0.11456082728567903,0.968
+7.0,0.06691350664702883,0.979625,0.10910866114092198,0.969
+8.0,0.059269082607526495,0.9825,0.10503831758585568,0.971
+9.0,0.053242110795707065,0.98375,0.10171043938325056,0.971
+10.0,0.04853800560203754,0.985125,0.09997651189261585,0.971
+11.0,0.044405441596568826,0.98575,0.09838001759817647,0.971
+12.0,0.040599887346456684,0.98725,0.097780955718731,0.972
+13.0,0.03734205810884943,0.98825,0.09713194222958135,0.974
+14.0,0.03431520433712973,0.98875,0.09692151531860237,0.974
+15.0,0.03161609671275515,0.98975,0.0973242406668757,0.976
+16.0,0.02932789356176544,0.990125,0.09678624020786661,0.974
+17.0,0.027317096210515864,0.990875,0.09601975313441609,0.975
+18.0,0.025400220809541275,0.99125,0.09493154670227145,0.975
+19.0,0.023849761916380875,0.992375,0.09431458156343589,0.975
+20.0,0.02235980249269527,0.992875,0.09444793299463022,0.975
+21.0,0.021040637629279574,0.993375,0.09406752804026423,0.976
+22.0,0.01981557992146146,0.99375,0.09397031761636573,0.976
+23.0,0.018703144106497697,0.994375,0.0936818870241683,0.976
+24.0,0.01755765508786901,0.995,0.09316670020081608,0.976
+25.0,0.01650424305425841,0.995375,0.09312778677582582,0.976
+26.0,0.015547213664301573,0.996,0.09313668531755637,0.976
+27.0,0.014641359094149346,0.99625,0.09262190181973926,0.975
+28.0,0.013849651147343996,0.996375,0.0924357805757844,0.975
+29.0,0.013045399464671547,0.99675,0.0917367938821814,0.974
+30.0,0.01230803442885487,0.996875,0.09119741077156332,0.974
+31.0,0.011604525329266091,0.997375,0.09061579658698077,0.973
+32.0,0.010950352940047849,0.997625,0.08979792577002656,0.973
+33.0,0.010404225458991357,0.997625,0.08958915531131897,0.973
+34.0,0.009935318745837484,0.99775,0.0894026029257118,0.973
+35.0,0.009447877430231724,0.99775,0.08878696599874977,0.973
+36.0,0.008965917587196202,0.997875,0.08823171309225875,0.974
+37.0,0.0085215831353927,0.99825,0.08802536709661077,0.973
+38.0,0.008150154843487975,0.998375,0.08802838908367926,0.973
+39.0,0.007784229897606622,0.9985,0.08738821100377808,0.973
+40.0,0.007436991028673095,0.9985,0.08739398645698664,0.973
+41.0,0.0071192159066914345,0.998625,0.08716156616155904,0.973
+42.0,0.006813014565763649,0.998625,0.08662144756300286,0.974
+43.0,0.0065184378060441145,0.99875,0.08661521759366482,0.975
+44.0,0.006218632412817654,0.999,0.08645243663927514,0.973
+45.0,0.005935930757015673,0.99925,0.08618233388001328,0.973
+46.0,0.005689173931263665,0.99925,0.08599448294668054,0.973
+47.0,0.005451752123017894,0.99925,0.08606674891635356,0.973
+48.0,0.005212159804542434,0.99925,0.08593790210578428,0.973
+49.0,0.004984321267572547,0.99925,0.08585662011359306,0.973
+50.0,0.004777153036591143,0.999375,0.08578410196981058,0.973
+51.0,0.0045963000083057756,0.999375,0.0860511141682166,0.973
+52.0,0.004398328316017485,0.999375,0.08587840406791315,0.974
+53.0,0.00420255185037797,0.9995,0.08593016802382229,0.974
+54.0,0.004025727702078051,0.9995,0.0856310412373703,0.974
+55.0,0.003846586557412409,0.999625,0.08534700213479356,0.975
+56.0,0.003690200877363095,0.99975,0.0853559191273964,0.975
+57.0,0.003528336774508031,0.999875,0.08520610384158768,0.975
+58.0,0.0033746694859510206,1.0,0.08517913142356824,0.976
+59.0,0.0032474977825223453,1.0,0.08501889024451946,0.976
+60.0,0.0031240067693776782,1.0,0.0848126655488919,0.977
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv.mtd
new file mode 100644
index 00000000000..bdad9be7c54
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-18 12:10:29 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv
new file mode 100644
index 00000000000..a0083b7d42b
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv
@@ -0,0 +1,90 @@
+1.0,2.3052508540430696,0.0977101337876018,2.303963296401079,0.1044
+2.0,2.303944458910519,0.0991955085590826,2.3035221142468574,0.1044
+3.0,2.3036721115786545,0.09878209655974737,2.303348806933503,0.0986
+4.0,2.3035195526911423,0.09792514957620076,2.3032560152519927,0.0986
+5.0,2.3034184316980806,0.097782325078943,2.303197726078531,0.0986
+6.0,2.3033450284556385,0.09803940917400697,2.303157331622366,0.0986
+7.0,2.303288583012619,0.0981536687718132,2.3031273807533745,0.0986
+8.0,2.303243403976425,0.09775376017949143,2.303104046162939,0.0986
+9.0,2.303206158991423,0.097782325078943,2.30308516976094,0.0986
+10.0,2.303174749851404,0.09729672178826658,2.303069447232022,0.0986
+11.0,2.3031477810136294,0.09675398869868705,2.3030560461012657,0.0986
+12.0,2.303124283577331,0.09681111849759015,2.3030444108621255,0.0986
+13.0,2.3031035611613118,0.09712533239155725,2.3030341567229007,0.0986
+14.0,2.3030850987567844,0.09763950058168522,2.303025008490508,0.0986
+15.0,2.3030685062423215,0.0975293958783447,2.303016763823304,0.0986
+16.0,2.3030534819916095,0.09704379258766828,2.30300927028908,0.0958
+17.0,2.3030397886431664,0.09621541050357321,2.3030024105684137,0.0958
+18.0,2.3030272365062925,0.09641536479973409,2.302996092635205,0.0958
+19.0,2.3030156719152237,0.09658675419644341,2.3029902430729923,0.0958
+20.0,2.3030049688788137,0.09715805218547449,2.3029848024211437,0.0958
+21.0,2.3029950229784286,0.09724374688382915,2.3029797218672385,0.0958
+22.0,2.3029857468327006,0.09701522768821673,2.3029749608519454,0.0958
+23.0,2.3029770666752962,0.09695809788931362,2.302970485304816,0.0958
+24.0,2.3029689197369243,0.09701522768821673,2.302966266324296,0.0958
+25.0,2.3029612522174836,0.09712948728602294,2.3029622791757705,0.0958
+26.0,2.302954017697477,0.0968438382915074,2.302958502520811,0.0958
+27.0,2.30294717588051,0.09681527339205585,2.3029549178169453,0.0958
+28.0,2.302940691588364,0.09698666278876517,2.3029515088448482,0.0958
+29.0,2.3029345339507907,0.09644392969918565,2.302948261331938,0.0958
+30.0,2.3029286757469483,0.09627254030247633,2.302945162649764,0.0958
+31.0,2.302923092866068,0.09601545620741234,2.3029422015684697,0.0958
+32.0,2.3029177638626317,0.09615828070467011,2.3029393680558727,0.0958
+33.0,2.302912669587062,0.09624397540302476,2.3029366531117326,0.0958
+34.0,2.3029077928772526,0.09675814359315274,2.3029340486300307,0.0958
+35.0,2.302903118299415,0.09681527339205585,2.302931547283736,0.0958
+36.0,2.302898631929166,0.09692953298986207,2.3029291424277525,0.0958
+37.0,2.302894321165697,0.09681527339205585,2.3029268280166884,0.0958
+38.0,2.3028901745732164,0.09755796077779624,2.3029245985347893,0.0958
+39.0,2.302886181745126,0.09741513628053848,2.3029224489359157,0.0958
+40.0,2.302882333187097,0.09738657138108693,2.302920374591868,0.0958
+41.0,2.3028786202160454,0.09744370117999003,2.3029183712476917,0.0958
+42.0,2.3028750348725207,0.0974722660794416,2.3029164349828504,0.0958
+43.0,2.302871569844405,0.0976436554761509,2.302914562177353,0.0958
+44.0,2.3028682184002816,0.09767222037560247,2.3029127494820916,0.0958
+45.0,2.3028649743310323,0.0975293958783447,2.3029109937927728,0.0958
+46.0,2.302861831898485,0.0975293958783447,2.3029092922269156,0.0958
+47.0,2.3028587857900877,0.09767222037560247,2.3029076421035097,0.0958
+48.0,2.302855831078908,0.09798643426956956,2.302906040924949,0.0958
+49.0,2.302852963188079,0.09804356406847267,2.302904486360953,0.0958
+50.0,2.3028501778592187,0.09801499916902111,2.302902976234211,0.0958
+51.0,2.3028474711242892,0.09847203756024597,2.3029015085075333,0.0958
+52.0,2.302844839280435,0.09855773225860064,2.302900081272322,0.0958
+53.0,2.3028422788674385,0.09844347266079442,2.3028986927381987,0.0958
+54.0,2.3028397866475188,0.09838634286189131,2.3028973412236535,0.0958
+55.0,2.3028373595871368,0.09858629715805219,2.3028960251475996,0.0958
+56.0,2.302834994840583,0.09844347266079442,2.302894743021725,0.0958
+57.0,2.3028326897351916,0.09847203756024597,2.30289349344355,0.0958
+58.0,2.302830441757941,0.09844347266079442,2.302892275090126,0.0958
+59.0,2.3028282485432734,0.09835777796243976,2.302891086712291,0.0958
+60.0,2.3028261078621295,0.09841490776134287,2.3028899271294314,0.0958
+61.0,2.3028240176118455,0.09844347266079442,2.3028887952246992,0.0958
+62.0,2.3028219758070656,0.09870055675585841,2.302887689940629,0.0958
+63.0,2.302819980571373,0.09867199185640685,2.3028866102751295,0.0958
+64.0,2.3028180301296777,0.09884338125311617,2.302885555277795,0.0958
+65.0,2.3028161228012136,0.09875768655476151,2.302884524046522,0.0958
+66.0,2.302814256993192,0.09878625145421306,2.3028835157243908,0.0958
+67.0,2.3028124311948646,0.09878625145421306,2.302882529496792,0.0958
+68.0,2.3028106439721734,0.09870055675585841,2.302881564588779,0.0958
+69.0,2.3028088939627684,0.09872912165530996,2.302880620262613,0.0958
+70.0,2.3028071798714174,0.09872912165530996,2.3028796958155,0.0958
+71.0,2.3028055004658077,0.09870055675585841,2.3028787905774917,0.0958
+72.0,2.302803854572614,0.09867199185640685,2.302877903909538,0.0958
+73.0,2.302802241073912,0.09884338125311617,2.302877035201682,0.0958
+74.0,2.302800658903833,0.09884338125311617,2.30287618387138,0.0958
+75.0,2.3027991070454563,0.0990147706498255,2.3028753493619405,0.0958
+76.0,2.302797584527948,0.0990147706498255,2.302874531141072,0.0958
+77.0,2.3027960904238864,0.0990147706498255,2.3028737286995273,0.0958
+78.0,2.3027946238467796,0.0990719004487286,2.302872941549839,0.0958
+79.0,2.302793183948759,0.0990719004487286,2.30287216922514,0.0958
+80.0,2.3027917699184237,0.09904333554927705,2.3028714112780633,0.0958
+81.0,2.3027903809788377,0.09904333554927705,2.3028706672797075,0.0958
+82.0,2.302789016385661,0.09892907595147084,2.3028699368186745,0.0958
+83.0,2.302787675425392,0.09895764085092239,2.302869219500162,0.0958
+84.0,2.302786357413738,0.09884338125311617,2.3028685149451196,0.0958
+85.0,2.3027850616940877,0.09875768655476151,2.3028678227894512,0.0958
+86.0,2.302783787636076,0.09847203756024597,2.302867142683268,0.0958
+87.0,2.3027825346342436,0.09838634286189131,2.3028664742901888,0.0958
+88.0,2.3027813021067707,0.09838634286189131,2.3028658172866763,0.0958
+89.0,2.3027800894943056,0.09838634286189131,2.302865171361421,0.0958
+90.0,2.3027788962588445,0.09847203756024597,2.302864536214754,0.0958
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv.mtd
new file mode 100644
index 00000000000..3337bcddcd5
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_cifar.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 90,
+    "cols": 5,
+    "nnz": 450,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 15:02:57 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv
new file mode 100644
index 00000000000..df2e0b9dec5
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv
@@ -0,0 +1,60 @@
+1.0,0.6395330965863911,0.8015,0.3272391477006809,0.902
+2.0,0.18475208110972252,0.945375,0.22124293746528123,0.937
+3.0,0.13424154347106543,0.95975,0.18851124248025902,0.941
+4.0,0.10030261909915536,0.970375,0.1667022281154376,0.951
+5.0,0.07937873593518888,0.976,0.15600958058409162,0.957
+6.0,0.06920171125717522,0.979125,0.16498212073116425,0.957
+7.0,0.060592194621787006,0.98125,0.17178759629686993,0.958
+8.0,0.05412933187594701,0.98275,0.1633657632581319,0.962
+9.0,0.04811218647175158,0.985375,0.1497524304739882,0.963
+10.0,0.042713873413571295,0.987625,0.13528177549249587,0.964
+11.0,0.03798921289828895,0.989,0.12257248782459493,0.966
+12.0,0.033368917980423234,0.990125,0.11437361989464159,0.968
+13.0,0.02928468918448111,0.991125,0.10654460042550556,0.973
+14.0,0.025635927828060606,0.99225,0.10001953244032351,0.971
+15.0,0.022313700189017967,0.993375,0.09378573210443951,0.97
+16.0,0.01957044397816946,0.99475,0.0884900055429675,0.973
+17.0,0.01725229577517209,0.996125,0.0840339954715082,0.976
+18.0,0.015548186647948697,0.99625,0.08210346019097332,0.976
+19.0,0.014139138023695867,0.996875,0.08060906399263776,0.976
+20.0,0.012959707879691243,0.99775,0.07868875058565629,0.976
+21.0,0.012004452751684604,0.9975,0.07724035363259926,0.977
+22.0,0.0111136327233965,0.9975,0.07612728983693134,0.977
+23.0,0.010281475186653103,0.998,0.07584562740464813,0.976
+24.0,0.009677989022521137,0.998,0.07425610274151832,0.976
+25.0,0.008997284040789153,0.998,0.0736737006721429,0.977
+26.0,0.008524000701748208,0.998,0.07296239630790295,0.977
+27.0,0.007973775129518497,0.99825,0.07233596593358264,0.976
+28.0,0.0075191435573975,0.998625,0.0717720466101364,0.975
+29.0,0.007066838280190324,0.99875,0.0702526716863536,0.974
+30.0,0.006566122743096105,0.999,0.06965647460292666,0.974
+31.0,0.006089250289720929,0.999,0.06819989827467632,0.977
+32.0,0.005608455593168797,0.999125,0.06694592070023135,0.977
+33.0,0.005141466669596634,0.99925,0.06595043354269733,0.978
+34.0,0.004759061968380032,0.99925,0.06443985064969614,0.978
+35.0,0.004380594774155363,0.999375,0.06353770184635456,0.978
+36.0,0.004046733048490616,0.99975,0.06253301596760709,0.979
+37.0,0.003786776513232984,0.99975,0.061590454895157895,0.979
+38.0,0.00349652858317629,1.0,0.0608157597997437,0.979
+39.0,0.003270031704071804,1.0,0.06014196945494181,0.979
+40.0,0.0030958052763965156,1.0,0.05944900177792484,0.979
+41.0,0.0029148209401834786,1.0,0.05910852235695886,0.979
+42.0,0.002747487564857834,1.0,0.05885799247405729,0.98
+43.0,0.002601885279135758,1.0,0.05875771036367837,0.98
+44.0,0.002452998785943756,1.0,0.05875073571192066,0.98
+45.0,0.0023251227606833993,1.0,0.05883290853287607,0.98
+46.0,0.002200477357490353,1.0,0.05931266420926622,0.98
+47.0,0.0020748584217780814,1.0,0.059344483349665564,0.98
+48.0,0.001972244440490512,1.0,0.059807717143902615,0.98
+49.0,0.001866432894600097,1.0,0.06028438945471628,0.98
+50.0,0.0017845509941932457,1.0,0.06072764247481106,0.98
+51.0,0.0017029988093170635,1.0,0.06114998420916026,0.98
+52.0,0.0016256411665798158,1.0,0.06155595808269873,0.979
+53.0,0.0015603666542434512,1.0,0.061922900447717436,0.979
+54.0,0.0014921414498039287,1.0,0.06227737736439052,0.979
+55.0,0.001441256508702496,1.0,0.06265306885356355,0.979
+56.0,0.0013807884045605036,1.0,0.06304211962247053,0.979
+57.0,0.0013330958389548527,1.0,0.063426241980221,0.979
+58.0,0.0012818933369036054,1.0,0.06378686691752658,0.979
+59.0,0.0012389867278221342,1.0,0.06408567297048824,0.979
+60.0,0.001195307205982342,1.0,0.06431335760551511,0.98
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv.mtd
new file mode 100644
index 00000000000..e3d62bb00c8
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_diagonal_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-17 12:21:48 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv
new file mode 100644
index 00000000000..475eacfbd8b
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv
@@ -0,0 +1,60 @@
+1.0,0.6395330965863912,0.8015,0.32723914770068085,0.902
+2.0,0.18475208110972235,0.945375,0.22124293746528062,0.937
+3.0,0.1342415434710652,0.95975,0.18851124248025777,0.941
+4.0,0.10030261909915518,0.970375,0.16670222811543603,0.951
+5.0,0.07937873593518883,0.976,0.15600958058409148,0.957
+6.0,0.06920171125717528,0.979125,0.1649821207311639,0.957
+7.0,0.06059219462178708,0.98125,0.17178759629687052,0.958
+8.0,0.05412933187594715,0.98275,0.16336576325813237,0.962
+9.0,0.048112186471751854,0.985375,0.1497524304739886,0.963
+10.0,0.04271387341357169,0.987625,0.13528177549249676,0.964
+11.0,0.037989212898289285,0.989,0.12257248782459636,0.966
+12.0,0.0333689179804235,0.990125,0.11437361989464215,0.968
+13.0,0.029284689184481228,0.991125,0.10654460042550568,0.973
+14.0,0.02563592782806072,0.99225,0.10001953244032429,0.971
+15.0,0.022313700189017974,0.993375,0.09378573210444012,0.97
+16.0,0.01957044397816949,0.99475,0.08849000554296813,0.973
+17.0,0.017252295775172097,0.996125,0.08403399547150898,0.976
+18.0,0.015548186647948742,0.99625,0.08210346019097452,0.976
+19.0,0.014139138023695867,0.996875,0.08060906399263904,0.976
+20.0,0.012959707879691218,0.99775,0.07868875058565691,0.976
+21.0,0.012004452751684557,0.9975,0.07724035363260007,0.977
+22.0,0.01111363272339649,0.9975,0.07612728983693215,0.977
+23.0,0.010281475186653124,0.998,0.07584562740464891,0.976
+24.0,0.009677989022521148,0.998,0.07425610274151902,0.976
+25.0,0.008997284040789163,0.998,0.07367370067214365,0.977
+26.0,0.008524000701748191,0.998,0.0729623963079035,0.977
+27.0,0.00797377512951854,0.99825,0.07233596593358348,0.976
+28.0,0.007519143557397548,0.998625,0.07177204661013718,0.975
+29.0,0.007066838280190385,0.99875,0.07025267168635428,0.974
+30.0,0.006566122743096165,0.999,0.06965647460292756,0.974
+31.0,0.006089250289720989,0.999,0.06819989827467741,0.977
+32.0,0.005608455593168826,0.999125,0.06694592070023227,0.977
+33.0,0.005141466669596642,0.99925,0.06595043354269849,0.978
+34.0,0.004759061968380081,0.99925,0.06443985064969704,0.978
+35.0,0.004380594774155382,0.999375,0.0635377018463556,0.978
+36.0,0.004046733048490658,0.99975,0.06253301596760812,0.979
+37.0,0.0037867765132330333,0.99975,0.06159045489515902,0.979
+38.0,0.003496528583176314,1.0,0.060815759799744844,0.979
+39.0,0.00327003170407184,1.0,0.060141969454942976,0.979
+40.0,0.0030958052763965277,1.0,0.05944900177792594,0.979
+41.0,0.0029148209401835042,1.0,0.05910852235696025,0.979
+42.0,0.0027474875648578626,1.0,0.05885799247405899,0.98
+43.0,0.0026018852791357774,1.0,0.05875771036368049,0.98
+44.0,0.002452998785943783,1.0,0.058750735711922415,0.98
+45.0,0.002325122760683419,1.0,0.058832908532878384,0.98
+46.0,0.0022004773574903613,1.0,0.05931266420926835,0.98
+47.0,0.0020748584217780996,1.0,0.059344483349667965,0.98
+48.0,0.001972244440490512,1.0,0.0598077171439049,0.98
+49.0,0.001866432894600093,1.0,0.060284389454718604,0.98
+50.0,0.001784550994193222,1.0,0.06072764247481326,0.98
+51.0,0.0017029988093170592,1.0,0.061149984209162594,0.98
+52.0,0.00162564116657979,1.0,0.061555958082701305,0.979
+53.0,0.0015603666542434351,1.0,0.061922900447719774,0.979
+54.0,0.0014921414498039198,1.0,0.062277377364392435,0.979
+55.0,0.001441256508702488,1.0,0.06265306885356571,0.979
+56.0,0.0013807884045604901,1.0,0.06304211962247282,0.979
+57.0,0.0013330958389548362,1.0,0.06342624198022304,0.979
+58.0,0.0012818933369035971,1.0,0.06378686691752886,0.979
+59.0,0.0012389867278221167,1.0,0.06408567297049027,0.979
+60.0,0.0011953072059823309,1.0,0.06431335760551778,0.98
diff --git a/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv.mtd b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv.mtd
new file mode 100644
index 00000000000..604f85a9c2e
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/metrics/metrics_shampoo_momentum_mnist.csv.mtd
@@ -0,0 +1,12 @@
+{
+    "data_type": "matrix",
+    "value_type": "double",
+    "rows": 60,
+    "cols": 5,
+    "nnz": 300,
+    "format": "csv",
+    "author": "nicol",
+    "header": false,
+    "sep": ",",
+    "created": "2026-01-18 12:19:24 MEZ"
+}
\ No newline at end of file
diff --git a/scripts/staging/shampoo_optimizer/shampoo_optimizer_experiments.dml b/scripts/staging/shampoo_optimizer/shampoo_optimizer_experiments.dml
new file mode 100644
index 00000000000..658724473eb
--- /dev/null
+++ b/scripts/staging/shampoo_optimizer/shampoo_optimizer_experiments.dml
@@ -0,0 +1,482 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/optim/shampoo.dml") as shampoo
+source("scripts/nn/optim/adagrad.dml") as adagrad
+source("scripts/nn/optim/adam.dml") as adam
+
+source("scripts/nn/layers/conv2d_builtin.dml") as conv2d
+source("scripts/nn/layers/avg_pool2d_builtin.dml") as avg_pool2d
+source("scripts/nn/layers/relu.dml") as relu
+source("scripts/nn/layers/cross_entropy_loss.dml") as cross_entropy_loss
+source("scripts/nn/layers/softmax.dml") as softmax
+
+source("src/test/scripts/applications/nn/util.dml") as test_util
+
+
+# defining forward pass
+modelPredict = function(matrix[double] X, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc, list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins)
+    return(matrix[double] softMaxOut, matrix[double] X, matrix[double] convOut1, matrix[double] poolOut1, matrix[double] reluOut1, matrix[double] convOut2, matrix[double] poolOut2, matrix[double] reluOut2, matrix[double] pred){ #
+        filters = 64
+        conv_kernel = 5
+        pool_kernel = 4
+        conv_padding = 2
+        pool_padding = 1
+
+        h_in = as.integer(as.scalar(h_ins[1]))
+        h_in_1 = as.integer(as.scalar(h_ins[2]))
+        h_in_2 = as.integer(as.scalar(h_ins[3]))
+        h_in_3 = as.integer(as.scalar(h_ins[4]))
+        h_in_4 = as.integer(as.scalar(h_ins[5]))
+
+        w_in = as.integer(as.scalar(w_ins[1]))
+        w_in_1 = as.integer(as.scalar(w_ins[2]))
+        w_in_2 = as.integer(as.scalar(w_ins[3]))
+        w_in_3 = as.integer(as.scalar(w_ins[4]))
+        w_in_4 = as.integer(as.scalar(w_ins[5]))
+
+        c_in = as.integer(as.scalar(c_ins[1]))
+        c_in_1 = as.integer(as.scalar(c_ins[2]))
+
+        # first block
+        [convOut1, Hout, Wout] = conv2d::forward(X, W1, b1, c_in, h_in, w_in, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        [poolOut1, Hout, Wout] = avg_pool2d::forward(convOut1, c_in_1, h_in_1, w_in_1, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        reluOut1 = relu::forward(poolOut1)
+
+        # second block
+        [convOut2, Hout, Wout] = conv2d::forward(reluOut1, W2, b2, c_in_1, h_in_2, w_in_2, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        [poolOut2, Hout, Wout] = avg_pool2d::forward(convOut2, c_in_1, h_in_3, w_in_3, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        reluOut2 = relu::forward(poolOut2)
+
+        pred = reluOut2 %*% t(W_fc) + t(b_fc)
+
+        softMaxOut = softmax::forward(pred)
+
+        #Xs = list(X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred, softMaxOut)
+
+    }
+
+#defining backward pass
+modelBackward = function(matrix[double] target, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc, list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins, matrix[double] X, matrix[double] convOut1, matrix[double] poolOut1, matrix[double] reluOut1, matrix[double] convOut2, matrix[double] poolOut2, matrix[double] reluOut2, matrix[double] pred, matrix[double] softMaxOut)
+    return (matrix[double] gradient_lin_layer_W, matrix[double] gradient_lin_layer_b, matrix[double] gradient_W2, matrix[double] gradient_b2, matrix[double] gradient_W1, matrix[double] gradient_b1){
+        filters = 64
+        conv_kernel = 5
+        pool_kernel = 4
+        conv_padding = 2
+        pool_padding = 1
+
+        h_in = as.integer(as.scalar(h_ins[1]))
+        h_in_1 = as.integer(as.scalar(h_ins[2]))
+        h_in_2 = as.integer(as.scalar(h_ins[3]))
+        h_in_3 = as.integer(as.scalar(h_ins[4]))
+        h_in_4 = as.integer(as.scalar(h_ins[5]))
+
+        w_in = as.integer(as.scalar(w_ins[1]))
+        w_in_1 = as.integer(as.scalar(w_ins[2]))
+        w_in_2 = as.integer(as.scalar(w_ins[3]))
+        w_in_3 = as.integer(as.scalar(w_ins[4]))
+        w_in_4 = as.integer(as.scalar(w_ins[5]))
+
+        c_in = as.integer(as.scalar(c_ins[1]))
+        c_in_1 = as.integer(as.scalar(c_ins[2]))
+
+        # gradient of loss function
+        gradient_lossfn = cross_entropy_loss::backward(softMaxOut, target)
+
+        # gradient Softmax
+        gradient_softmax = softmax::backward(gradient_lossfn, pred)
+
+        # gradients Linear layer
+        gradient_lin_layer_W = t(gradient_softmax) %*% reluOut2
+        gradient_lin_layer_b = t(colSums(gradient_softmax))
+        gradient_lin_layer_X = gradient_softmax %*% W_fc
+
+        # gradient second Relu
+        gradient_relu = relu::backward(gradient_lin_layer_X, poolOut2)
+
+        # gradient second pooling layer
+        gradient_second_pooling = avg_pool2d::backward(gradient_relu, h_in_4, w_in_4, convOut2, c_in_1, h_in_3, w_in_3, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        # gradient second conv layer
+        [gradient_second_conv_X, gradient_W2, gradient_b2] = conv2d::backward(gradient_second_pooling, h_in_3, w_in_3, reluOut1, W2, b2, c_in_1, h_in_2, w_in_2, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        # gradient of the first Relu
+        gradient_relu = relu::backward(gradient_second_conv_X, poolOut1)
+
+        # gradient first pooling layer
+        gradient_first_pooling = avg_pool2d::backward(gradient_relu, h_in_2, w_in_2, convOut1, c_in_1, h_in_1, w_in_1, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        # gradient first conv layer
+        [gradient_first_conv_X, gradient_W1, gradient_b1] = conv2d::backward(gradient_first_pooling, h_in_1, w_in_1, X, W1, b1, c_in, h_in, w_in, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+    }
+
+
+definingData = function(string dataset_name)
+    return(matrix[double] X_train, matrix[double] Y_train, matrix[double] X_val, matrix[double] Y_val, matrix[double] X_test, matrix[double] Y_test){
+    if (dataset_name=="mnist"){
+        data = read("src/test/resources/datasets/MNIST/mnist_test.csv", format="csv") 
+        train = data[1:8999,]
+        test = data[9000:nrow(data),] 
+
+        images = train[,2:ncol(train)]
+        images = images / 255.0
+        labels = train[,1]
+        images_test = test[,2:ncol(test)]
+        labels_test = test[,1]
+
+        N = nrow(images)
+        N_test = nrow(images_test)     
+
+        X_train = images[1001:nrow(images),]
+        labels_train = labels[1001:nrow(images),]
+        Y_train = table(seq(1, nrow(X_train)), labels_train+1, nrow(X_train), 10)      
+
+        X_val = images[1:1000,]
+        labels_val = labels[1:1000,]
+        Y_val = table(seq(1, nrow(X_val)), labels_val+1, nrow(X_val), 10)
+
+        X_test = images_test
+        Y_test = table(seq(1, N_test), labels_test+1, N_test, 10)
+    }
+    if (dataset_name=="cifar"){
+        data = read("scripts/staging/shampoo_optimizer/cifar10.csv", format="csv")
+        train = data[1:39999,]                              
+        test = data[40000:nrow(data),]  
+
+        images = train[,2:ncol(train)]
+        images = images / 255.0
+        labels = train[,1]
+        images_test = test[,2:ncol(test)]
+        labels_test = test[,1]
+
+        N = nrow(images)
+        N_test = nrow(images_test)
+
+        X_train = images[5001:nrow(images),]                  
+        labels_train = labels[5001:nrow(images),] 
+        Y_train = table(seq(1, nrow(X_train)), labels_train+1, nrow(X_train), 10)
+
+        X_val = images[1:5000,]          
+        labels_val = labels[1:5000,] 
+        Y_val = table(seq(1, nrow(X_val)), labels_val+1, nrow(X_val), 10)
+
+        X_test = images_test
+        Y_test = table(seq(1, N_test), labels_test+1, N_test, 10)
+    }
+}
+
+# Define image properties
+
+defining_image_properties = function(string dataset_name)
+    return(int h_in, int w_in, int c_in, int classes){
+    if(dataset_name=="mnist"){
+        h_in = 28       
+        w_in = 28       
+        c_in = 1        
+        classes = 10
+    } 
+    if(dataset_name=="cifar"){
+        h_in = 32       
+        w_in = 32      
+        c_in = 3        
+        classes = 10
+    }
+}
+
+# Define training parameters
+defining_training_parameters = function(string optimizer)
+    return(int epochs, int batch_size, double epsilon, double lr, int diagThreshold, int rootEvery, int preconEvery){
+    if(optimizer=="adam"){
+        epsilon = 1e-8
+        lr = 0.001         
+    } 
+    else if((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+        epsilon = 1e-5
+        lr = 0.005       
+    } 
+    else{
+        epsilon = 1e-3 
+        lr = 0.6
+    }
+    epochs = 60 
+    batch_size = 64
+    diagThreshold = 1200
+    rootEvery = 10  
+    preconEvery = 2
+}
+
+defining_model_parameters = function()
+    return(int filters, int conv_kernel, int pool_kernel, int conv_padding, int pool_padding, int seed){
+    
+    filters = 64
+    conv_kernel = 5
+    pool_kernel = 4
+    conv_padding = 2
+    pool_padding = 1
+    seed = 42    
+}
+
+# create simple nn for image classification
+defining_nn_image_classification = function(int h_in, int w_in, int c_in, int classes, int filters, int conv_kernel, int pool_kernel, int conv_padding, int pool_padding, int seed)
+    return(list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc){
+
+    # convolution layer 1
+    [W1, b1] = conv2d::init(filters, c_in, conv_kernel, conv_kernel, seed)
+    h_in_1 = h_in + conv_padding*2 - (conv_kernel - 1)
+    w_in_1 = w_in + conv_padding*2 - (conv_kernel - 1)
+    c_in_1 = filters
+    # pooling
+    h_in_2 = floor((h_in_1 + pool_padding*2 - pool_kernel)/2)+1
+    w_in_2 = floor((w_in_1 + pool_padding*2 - pool_kernel)/2)+1
+    # relu
+
+    # convolution layer 2
+    [W2, b2] = conv2d::init(filters, c_in_1, conv_kernel, conv_kernel, seed)
+    h_in_3 = h_in_2 + conv_padding*2 - (conv_kernel - 1)
+    w_in_3 = w_in_2 + conv_padding*2 - (conv_kernel - 1)
+    c_in_1 = filters
+    # pooling
+    h_in_4 = floor((h_in_3 + pool_padding*2 - pool_kernel)/2)+1
+    w_in_4 = floor((w_in_3 + pool_padding*2 - pool_kernel)/2)+1
+    # relu
+
+    # Linear
+    W_fc = rand(rows=classes, cols=h_in_4*w_in_4*c_in_1, pdf="uniform", min=-0.1, max=0.1, seed=seed)
+    b_fc = matrix(0, rows=classes, cols=1)
+
+    h_ins = list(h_in, h_in_1, h_in_2, h_in_3, h_in_4)
+    w_ins = list(w_in, w_in_1, w_in_2, w_in_3, w_in_4)
+    c_ins = list(c_in, c_in_1)    
+}
+
+get_optimizer = function(int optimizer_index)
+    return(string optimizer){
+
+        if (optimizer_index==4)
+        {
+            optimizer = "adam"
+        }
+        if (optimizer_index==1)
+        {
+            optimizer = "shampoo"
+        }
+        if (optimizer_index==5)
+        {
+            optimizer = "shampoo_diagonal"
+        }
+        if (optimizer_index==2)
+        {
+            optimizer = "shampoo_momentum"
+        }
+        if (optimizer_index==6)
+        {
+            optimizer = "shampoo_momentum_diagonal"
+        }
+        if (optimizer_index==3)
+        {
+            optimizer = "shampoo_heuristic"
+        }
+        if (optimizer_index==7)
+        {
+            optimizer = "shampoo_heuristic_diagonal"
+        }
+}
+
+# set parameters for the experiments
+#############################################################################################
+dataset_name = "mnist" #Alternatives: "mnist" or "cifar"
+optimizers_to_experiment = list("shampoo", "shampoo_momentum", "shampoo_heuristic", "adam")
+# Alternatives: ("shampoo", "shampoo_diagonal", "shampoo_momentum", 
+#           "shampoo_momentum_diagonal", "shampoo_heuristic", "shampoo_heuristic_diagonal", "adam")
+#############################################################################################
+
+for (optimizer_index in 1:length(optimizers_to_experiment)){
+    optimizer = get_optimizer(optimizer_index)
+    print("Starting with " + optimizer)
+
+    # get the data
+    [X_train, Y_train, X_val, Y_val, X_test, Y_test] = definingData(dataset_name)
+
+    # get image properties
+    [h_in, w_in, c_in, classes] = defining_image_properties(dataset_name)
+
+    # get model parameters
+    [filters, conv_kernel, pool_kernel, conv_padding, pool_padding, seed] = defining_model_parameters()
+
+    #get model weights
+    [h_ins, w_ins, c_ins, W1, b1, W2, b2, W_fc, b_fc] = defining_nn_image_classification(h_in, w_in, c_in, classes, filters, conv_kernel, pool_kernel, conv_padding, pool_padding, seed)
+
+    # get training parameters
+    [epochs, batch_size, epsilon, lr, diagThreshold, rootEvery, preconEvery]= defining_training_parameters(optimizer)
+
+
+    if ((optimizer == "shampoo") | (optimizer == "shampoo_diagonal")){
+        [preconL_W1, preconR_W1, useDiag_W1] = shampoo::init(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, useDiag_b1] = shampoo::init(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, useDiag_W2] = shampoo::init(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, useDiag_b2] = shampoo::init(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, useDiag_W_fc] = shampoo::init(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, useDiag_b_fc] = shampoo::init(b_fc, epsilon, diagThreshold)
+    }
+    if ((optimizer == "shampoo_momentum") | (optimizer == "shampoo_momentum_diagonal")){
+        [preconL_W1, preconR_W1, momentum_W1, useDiag_W1] = shampoo::init_momentum(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, momentum_b1, useDiag_b1] = shampoo::init_momentum(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, momentum_W2, useDiag_W2] = shampoo::init_momentum(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, momentum_b2, useDiag_b2] = shampoo::init_momentum(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, momentum_W_fc, useDiag_W_fc] = shampoo::init_momentum(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, momentum_b_fc, useDiag_b_fc] = shampoo::init_momentum(b_fc, epsilon, diagThreshold)
+    }
+    if ((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+        [preconL_W1, preconR_W1, stepCounter_W1, bufferL_W1, bufferR_W1, momentum_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1, useDiag_W1] = shampoo::init_heuristic(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, stepCounter_b1, bufferL_b1, bufferR_b1, momentum_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1, useDiag_b1] = shampoo::init_heuristic(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, stepCounter_W2, bufferL_W2, bufferR_W2, momentum_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2, useDiag_W2] = shampoo::init_heuristic(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, stepCounter_b2, bufferL_b2, bufferR_b2, momentum_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2, useDiag_b2] = shampoo::init_heuristic(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, stepCounter_W_fc, bufferL_W_fc, bufferR_W_fc, momentum_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc, useDiag_W_fc] = shampoo::init_heuristic(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, stepCounter_b_fc, bufferL_b_fc, bufferR_b_fc, momentum_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc, useDiag_b_fc] = shampoo::init_heuristic(b_fc, epsilon, diagThreshold)
+    }
+    if (optimizer == "adam"){
+        [m_W1, v_W1] = adam::init(W1)
+        [m_b1, v_b1] = adam::init(b1)
+        [m_W2, v_W2] = adam::init(W2)
+        [m_b2, v_b2] = adam::init(b2)
+        [m_W_fc, v_W_fc] = adam::init(W_fc)
+        [m_b_fc, v_b_fc] = adam::init(b_fc)
+    }
+
+    data_val_X = X_val
+    data_val_Y = Y_val
+
+    # define the training
+
+    train_losses = matrix(0, rows=epochs, cols=1)
+    train_accuracies = matrix(0, rows=epochs, cols=1)
+    val_accuracies = matrix(0, rows=epochs, cols=1)
+    val_losses = matrix(0, rows=epochs, cols=1)
+    Ntrain = nrow(X_train)
+
+    timestep = 0
+
+    for(epoch in 1:epochs){
+
+        print("Epoch " + epoch + " of " + epochs + " epochs")
+
+        accuracy_value = 0
+        accuracy_count = 0
+        loss_value = 0
+        loss_count = 0
+        
+
+        for(start_index in seq(1, Ntrain, batch_size)){
+            #start_index = (i - 1) * batch_size + 1
+            end_index = min(start_index + batch_size - 1, Ntrain)
+            data_train_X = X_train[start_index:end_index,]
+            data_train_Y = Y_train[start_index:end_index,]
+
+            [softMaxOut, X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred] = modelPredict(data_train_X, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+            predicted_value = rowIndexMax(softMaxOut) - 1
+            accuracy = sum(predicted_value==rowIndexMax(data_train_Y)-1) / length(predicted_value)
+            accuracy_value = accuracy_value + accuracy
+            accuracy_count = accuracy_count + 1
+
+            loss = cross_entropy_loss::forward(softMaxOut, data_train_Y)
+            loss_value = loss_value + loss
+            loss_count = loss_count + 1
+            
+
+            [gradient_lin_layer_W, gradient_lin_layer_b, gradient_W2, gradient_b2, gradient_W1, gradient_b1] = modelBackward(data_train_Y, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins, X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred, softMaxOut)
+
+            if ((optimizer == "shampoo") | (optimizer == "shampoo_diagonal")){
+                [W1, preconL_W1, preconR_W1] = shampoo::update(W1, gradient_W1, lr, preconL_W1, preconR_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1] = shampoo::update(b1, gradient_b1, lr, preconL_b1, preconR_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2] = shampoo::update(W2, gradient_W2, lr, preconL_W2, preconR_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2] = shampoo::update(b2, gradient_b2, lr, preconL_b2, preconR_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc] = shampoo::update(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc] = shampoo::update(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, useDiag_b_fc)
+            }
+            if ((optimizer == "shampoo_momentum") | (optimizer == "shampoo_momentum_diagonal")){
+                [W1, preconL_W1, preconR_W1, momentum_W1] = shampoo::update_momentum(W1, gradient_W1, lr, preconL_W1, preconR_W1, momentum_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1, momentum_b1] = shampoo::update_momentum(b1, gradient_b1, lr, preconL_b1, preconR_b1, momentum_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2, momentum_W2] = shampoo::update_momentum(W2, gradient_W2, lr, preconL_W2, preconR_W2, momentum_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2, momentum_b2] = shampoo::update_momentum(b2, gradient_b2, lr, preconL_b2, preconR_b2, momentum_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc, momentum_W_fc] = shampoo::update_momentum(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, momentum_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc, momentum_b_fc] = shampoo::update_momentum(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, momentum_b_fc, useDiag_b_fc)
+            }
+            if ((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+                [W1, preconL_W1, preconR_W1, momentum_W1, stepCounter_W1, bufferL_W1, bufferR_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1] = shampoo::update_heuristic(W1, gradient_W1, lr, preconL_W1, preconR_W1, momentum_W1, stepCounter_W1, rootEvery, preconEvery, bufferL_W1, bufferR_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1, momentum_b1, stepCounter_b1, bufferL_b1, bufferR_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1] = shampoo::update_heuristic(b1, gradient_b1, lr, preconL_b1, preconR_b1, momentum_b1, stepCounter_b1, rootEvery, preconEvery, bufferL_b1, bufferR_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2, momentum_W2, stepCounter_W2, bufferL_W2, bufferR_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2] = shampoo::update_heuristic(W2, gradient_W2, lr, preconL_W2, preconR_W2, momentum_W2, stepCounter_W2, rootEvery, preconEvery, bufferL_W2, bufferR_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2, momentum_b2, stepCounter_b2, bufferL_b2, bufferR_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2] = shampoo::update_heuristic(b2, gradient_b2, lr, preconL_b2, preconR_b2, momentum_b2, stepCounter_b2, rootEvery, preconEvery, bufferL_b2, bufferR_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc, momentum_W_fc, stepCounter_W_fc, bufferL_W_fc, bufferR_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc] = shampoo::update_heuristic(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, momentum_W_fc, stepCounter_W_fc, rootEvery, preconEvery, bufferL_W_fc, bufferR_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc, momentum_b_fc, stepCounter_b_fc, bufferL_b_fc, bufferR_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc] = shampoo::update_heuristic(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, momentum_b_fc, stepCounter_b_fc, rootEvery, preconEvery, bufferL_b_fc, bufferR_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc, useDiag_b_fc)
+            }
+            if (optimizer == "adam"){
+                [W1, m_W1, v_W1] = adam::update(W1, gradient_W1, lr, 0.9, 0.999, epsilon, timestep, m_W1, v_W1)
+                [b1, m_b1, v_b1] = adam::update(b1, gradient_b1, lr, 0.9, 0.999, epsilon, timestep, m_b1, v_b1)
+                [W2, m_W2, v_W2] = adam::update(W2, gradient_W2, lr, 0.9, 0.999, epsilon, timestep, m_W2, v_W2)
+                [b2, m_b2, v_b2] = adam::update(b2, gradient_b2, lr, 0.9, 0.999, epsilon, timestep, m_b2, v_b2)
+                [W_fc, m_W_fc, v_W_fc] = adam::update(W_fc, gradient_lin_layer_W, lr, 0.9, 0.999, epsilon, timestep, m_W_fc, v_W_fc)
+                [b_fc, m_b_fc, v_b_fc] = adam::update(b_fc, gradient_lin_layer_b, lr, 0.9, 0.999, epsilon, timestep, m_b_fc, v_b_fc)
+                timestep = timestep + 1
+            }
+        }
+
+        train_losses[epoch,1] = loss_value / loss_count
+        train_accuracies[epoch,1] = accuracy_value/accuracy_count
+
+        [softMaxOut_val, X_val, convOut1_val, poolOut1_val, reluOut1_val, convOut2_val, poolOut2_val, reluOut2_val, pred_val] = modelPredict(X_val, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+
+        predicted_value_val = rowIndexMax(softMaxOut_val) - 1
+        accuracy_val = sum(predicted_value_val==rowIndexMax(Y_val)-1) / length(predicted_value_val)
+
+        val_accuracies[epoch,1] = accuracy_val
+
+        loss = cross_entropy_loss::forward(softMaxOut_val, Y_val)
+        val_losses[epoch,1] = loss
+    }
+
+
+    # define the testing
+
+    [softMaxOut_test, X_test, convOut1_test, poolOut1_test, reluOut1_test, convOut2_test, poolOut2_test, reluOut2_test, pred_test] = modelPredict(X_test, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+    predicted_value = rowIndexMax(softMaxOut_test) - 1
+    accuracy = sum(predicted_value==rowIndexMax(Y_test)-1) / length(predicted_value)
+
+    loss = cross_entropy_loss::forward(softMaxOut_test, Y_test)
+
+    outDir = "scripts/staging/shampoo_optimizer/metrics"
+    epochs = nrow(train_losses)
+    epoch_col = seq(1, epochs)         
+    M = cbind(epoch_col, train_losses, train_accuracies, val_losses, val_accuracies)
+    write(M, outDir + "/metrics" + "_" + optimizer + "_" + dataset_name + ".csv", format="csv")
+
+    print("Test Accuracy of " + optimizer + " on " + dataset_name + " = " + accuracy)
+    print("Test Loss of " + optimizer + " on " + dataset_name + " = " + loss)
+
+}
+
diff --git a/src/test/scripts/applications/nn/component/shampoo_test.dml b/src/test/scripts/applications/nn/component/shampoo_test.dml
new file mode 100644
index 00000000000..2ed90845261
--- /dev/null
+++ b/src/test/scripts/applications/nn/component/shampoo_test.dml
@@ -0,0 +1,437 @@
+#-------------------------------------------------------------
+#
+# Licensed to the Apache Software Foundation (ASF) under one
+# or more contributor license agreements.  See the NOTICE file
+# distributed with this work for additional information
+# regarding copyright ownership.  The ASF licenses this file
+# to you under the Apache License, Version 2.0 (the
+# "License"); you may not use this file except in compliance
+# with the License.  You may obtain a copy of the License at
+#
+#   http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing,
+# software distributed under the License is distributed on an
+# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+# KIND, either express or implied.  See the License for the
+# specific language governing permissions and limitations
+# under the License.
+#
+#-------------------------------------------------------------
+
+source("scripts/nn/optim/shampoo.dml") as shampoo
+source("scripts/nn/optim/adagrad.dml") as adagrad
+source("scripts/nn/optim/adam.dml") as adam
+
+source("scripts/nn/layers/conv2d_builtin.dml") as conv2d
+source("scripts/nn/layers/avg_pool2d_builtin.dml") as avg_pool2d
+source("scripts/nn/layers/relu.dml") as relu
+source("scripts/nn/layers/cross_entropy_loss.dml") as cross_entropy_loss
+source("scripts/nn/layers/softmax.dml") as softmax
+
+
+
+# defining forward pass
+modelPredict = function(matrix[double] X, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc, list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins)
+    return(matrix[double] softMaxOut, matrix[double] X, matrix[double] convOut1, matrix[double] poolOut1, matrix[double] reluOut1, matrix[double] convOut2, matrix[double] poolOut2, matrix[double] reluOut2, matrix[double] pred){ #
+        filters = 64
+        conv_kernel = 5
+        pool_kernel = 4
+        conv_padding = 2
+        pool_padding = 1
+
+        h_in = as.integer(as.scalar(h_ins[1]))
+        h_in_1 = as.integer(as.scalar(h_ins[2]))
+        h_in_2 = as.integer(as.scalar(h_ins[3]))
+        h_in_3 = as.integer(as.scalar(h_ins[4]))
+        h_in_4 = as.integer(as.scalar(h_ins[5]))
+
+        w_in = as.integer(as.scalar(w_ins[1]))
+        w_in_1 = as.integer(as.scalar(w_ins[2]))
+        w_in_2 = as.integer(as.scalar(w_ins[3]))
+        w_in_3 = as.integer(as.scalar(w_ins[4]))
+        w_in_4 = as.integer(as.scalar(w_ins[5]))
+
+        c_in = as.integer(as.scalar(c_ins[1]))
+        c_in_1 = as.integer(as.scalar(c_ins[2]))
+
+        # first block
+        [convOut1, Hout, Wout] = conv2d::forward(X, W1, b1, c_in, h_in, w_in, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        [poolOut1, Hout, Wout] = avg_pool2d::forward(convOut1, c_in_1, h_in_1, w_in_1, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        reluOut1 = relu::forward(poolOut1)
+
+        # second block
+        [convOut2, Hout, Wout] = conv2d::forward(reluOut1, W2, b2, c_in_1, h_in_2, w_in_2, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        [poolOut2, Hout, Wout] = avg_pool2d::forward(convOut2, c_in_1, h_in_3, w_in_3, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        reluOut2 = relu::forward(poolOut2)
+
+        pred = reluOut2 %*% t(W_fc) + t(b_fc)
+
+        softMaxOut = softmax::forward(pred)
+
+        #Xs = list(X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred, softMaxOut)
+
+    }
+
+#defining backward pass
+modelBackward = function(matrix[double] target, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc, list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins, matrix[double] X, matrix[double] convOut1, matrix[double] poolOut1, matrix[double] reluOut1, matrix[double] convOut2, matrix[double] poolOut2, matrix[double] reluOut2, matrix[double] pred, matrix[double] softMaxOut)
+    return (matrix[double] gradient_lin_layer_W, matrix[double] gradient_lin_layer_b, matrix[double] gradient_W2, matrix[double] gradient_b2, matrix[double] gradient_W1, matrix[double] gradient_b1){
+        filters = 64
+        conv_kernel = 5
+        pool_kernel = 4
+        conv_padding = 2
+        pool_padding = 1
+
+        h_in = as.integer(as.scalar(h_ins[1]))
+        h_in_1 = as.integer(as.scalar(h_ins[2]))
+        h_in_2 = as.integer(as.scalar(h_ins[3]))
+        h_in_3 = as.integer(as.scalar(h_ins[4]))
+        h_in_4 = as.integer(as.scalar(h_ins[5]))
+
+        w_in = as.integer(as.scalar(w_ins[1]))
+        w_in_1 = as.integer(as.scalar(w_ins[2]))
+        w_in_2 = as.integer(as.scalar(w_ins[3]))
+        w_in_3 = as.integer(as.scalar(w_ins[4]))
+        w_in_4 = as.integer(as.scalar(w_ins[5]))
+
+        c_in = as.integer(as.scalar(c_ins[1]))
+        c_in_1 = as.integer(as.scalar(c_ins[2]))
+
+        # gradient of loss function
+        gradient_lossfn = cross_entropy_loss::backward(softMaxOut, target)
+
+        # gradient Softmax
+        gradient_softmax = softmax::backward(gradient_lossfn, pred)
+
+        # gradients Linear layer
+        gradient_lin_layer_W = t(gradient_softmax) %*% reluOut2
+        gradient_lin_layer_b = t(colSums(gradient_softmax))
+        gradient_lin_layer_X = gradient_softmax %*% W_fc
+
+        # gradient second Relu
+        gradient_relu = relu::backward(gradient_lin_layer_X, poolOut2)
+
+        # gradient second pooling layer
+        gradient_second_pooling = avg_pool2d::backward(gradient_relu, h_in_4, w_in_4, convOut2, c_in_1, h_in_3, w_in_3, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        # gradient second conv layer
+        [gradient_second_conv_X, gradient_W2, gradient_b2] = conv2d::backward(gradient_second_pooling, h_in_3, w_in_3, reluOut1, W2, b2, c_in_1, h_in_2, w_in_2, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+        # gradient of the first Relu
+        gradient_relu = relu::backward(gradient_second_conv_X, poolOut1)
+
+        # gradient first pooling layer
+        gradient_first_pooling = avg_pool2d::backward(gradient_relu, h_in_2, w_in_2, convOut1, c_in_1, h_in_1, w_in_1, pool_kernel, pool_kernel, 2, 2, pool_padding, pool_padding)
+
+        # gradient first conv layer
+        [gradient_first_conv_X, gradient_W1, gradient_b1] = conv2d::backward(gradient_first_pooling, h_in_1, w_in_1, X, W1, b1, c_in, h_in, w_in, conv_kernel, conv_kernel, 1, 1, conv_padding, conv_padding)
+
+    }
+
+
+definingData = function()
+    return(matrix[double] X_train, matrix[double] Y_train, matrix[double] X_val, matrix[double] Y_val, matrix[double] X_test, matrix[double] Y_test){
+
+    data = read("src/test/resources/datasets/MNIST/mnist_test.csv", format="csv") 
+    train = data[1:8999,]
+    test = data[9000:nrow(data),] 
+
+    images = train[,2:ncol(train)]
+    images = images / 255.0
+    labels = train[,1]
+    images_test = test[,2:ncol(test)]
+    labels_test = test[,1]
+
+    N = nrow(images)
+    N_test = nrow(images_test)     
+
+    X_train = images[1001:nrow(images),]
+    labels_train = labels[1001:nrow(images),]
+    Y_train = table(seq(1, nrow(X_train)), labels_train+1, nrow(X_train), 10)      
+
+    X_val = images[1:1000,]
+    labels_val = labels[1:1000,]
+    Y_val = table(seq(1, nrow(X_val)), labels_val+1, nrow(X_val), 10)
+
+    X_test = images_test
+    Y_test = table(seq(1, N_test), labels_test+1, N_test, 10)
+}
+
+# Define image properties
+
+defining_image_properties = function()
+    return(int h_in, int w_in, int c_in, int classes){
+
+    h_in = 28       
+    w_in = 28       
+    c_in = 1        
+    classes = 10
+
+}
+
+# Define training parameters
+defining_training_parameters = function(string optimizer)
+    return(int epochs, int batch_size, double epsilon, double lr, int rootEvery, int preconEvery){
+    
+    if ((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+        epochs = 30
+        batch_size = 64
+        epsilon = 1e-4
+        lr = 0.005 
+        rootEvery = 10  
+        preconEvery = 2
+    }else{
+        epochs = 30
+        batch_size = 64
+        epsilon = 1e-4
+        lr = 0.005 
+        rootEvery = 0  
+        preconEvery = 0
+    }
+}
+
+defining_model_parameters = function()
+    return(int filters, int conv_kernel, int pool_kernel, int conv_padding, int pool_padding, int seed){
+    
+    filters = 64
+    conv_kernel = 5
+    pool_kernel = 4
+    conv_padding = 2
+    pool_padding = 1
+    seed = 42    
+}
+
+# create simple nn for image classification
+defining_nn_image_classification = function(int h_in, int w_in, int c_in, int classes, int filters, int conv_kernel, int pool_kernel, int conv_padding, int pool_padding, int seed)
+    return(list[unknown] h_ins, list[unknown] w_ins, list[unknown] c_ins, matrix[double] W1, matrix[double] b1, matrix[double] W2, matrix[double] b2, matrix[double] W_fc, matrix[double] b_fc){
+
+    # convolution layer 1
+    [W1, b1] = conv2d::init(filters, c_in, conv_kernel, conv_kernel, seed)
+    h_in_1 = h_in + conv_padding*2 - (conv_kernel - 1)
+    w_in_1 = w_in + conv_padding*2 - (conv_kernel - 1)
+    c_in_1 = filters
+    # pooling
+    h_in_2 = floor((h_in_1 + pool_padding*2 - pool_kernel)/2)+1
+    w_in_2 = floor((w_in_1 + pool_padding*2 - pool_kernel)/2)+1
+    # relu
+
+    # convolution layer 2
+    [W2, b2] = conv2d::init(filters, c_in_1, conv_kernel, conv_kernel, seed)
+    h_in_3 = h_in_2 + conv_padding*2 - (conv_kernel - 1)
+    w_in_3 = w_in_2 + conv_padding*2 - (conv_kernel - 1)
+    c_in_1 = filters
+    # pooling
+    h_in_4 = floor((h_in_3 + pool_padding*2 - pool_kernel)/2)+1
+    w_in_4 = floor((w_in_3 + pool_padding*2 - pool_kernel)/2)+1
+    # relu
+
+    # Linear
+    W_fc = rand(rows=classes, cols=h_in_4*w_in_4*c_in_1, pdf="uniform", min=-0.1, max=0.1, seed=seed)
+    b_fc = matrix(0, rows=classes, cols=1)
+
+    h_ins = list(h_in, h_in_1, h_in_2, h_in_3, h_in_4)
+    w_ins = list(w_in, w_in_1, w_in_2, w_in_3, w_in_4)
+    c_ins = list(c_in, c_in_1)    
+}
+
+get_optimizer = function(int optimizer_index)
+    return(string optimizer){
+
+        if (optimizer_index==1)
+        {
+            optimizer = "shampoo"
+        }
+        if (optimizer_index==2)
+        {
+            optimizer = "shampoo_diagonal"
+        }
+        if (optimizer_index==3)
+        {
+            optimizer = "shampoo_momentum"
+        }
+       if (optimizer_index==4)
+        {
+            optimizer = "shampoo_momentum_diagonal"
+        }
+        if (optimizer_index==5)
+        {
+            optimizer = "shampoo_heuristic"
+        }
+        if (optimizer_index==6)
+        {
+            optimizer = "shampoo_heuristic_diagonal"
+        }
+}
+
+# set parameters for the tests
+#############################################################################################
+optimizers_to_experiment = list("shampoo", "shampoo_diagonal", "shampoo_momentum", 
+            "shampoo_momentum_diagonal", "shampoo_heuristic", "shampoo_heuristic_diagonal")
+#############################################################################################
+
+for (optimizer_index in 1:length(optimizers_to_experiment)){
+    optimizer = get_optimizer(optimizer_index)
+    print("Starting with " + optimizer)
+
+    # get the data
+    [X_train, Y_train, X_val, Y_val, X_test, Y_test] = definingData()
+
+    # get image properties
+    [h_in, w_in, c_in, classes] = defining_image_properties()
+
+    # get model parameters
+    [filters, conv_kernel, pool_kernel, conv_padding, pool_padding, seed] = defining_model_parameters()
+
+    #get model weights
+    [h_ins, w_ins, c_ins, W1, b1, W2, b2, W_fc, b_fc] = defining_nn_image_classification(h_in, w_in, c_in, classes, filters, conv_kernel, pool_kernel, conv_padding, pool_padding, seed)
+
+    # get training parameters
+    [epochs, batch_size, epsilon, lr, rootEvery, preconEvery]= defining_training_parameters(optimizer)
+
+
+    if ((optimizer == "shampoo") | (optimizer == "shampoo_momentum") | (optimizer == "shampoo_heuristic")){
+        diagThreshold = 1200
+    }else{
+        diagThreshold = 1
+    }
+
+    if ((optimizer == "shampoo") | (optimizer == "shampoo_diagonal")){
+        [preconL_W1, preconR_W1, useDiag_W1] = shampoo::init(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, useDiag_b1] = shampoo::init(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, useDiag_W2] = shampoo::init(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, useDiag_b2] = shampoo::init(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, useDiag_W_fc] = shampoo::init(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, useDiag_b_fc] = shampoo::init(b_fc, epsilon, diagThreshold)
+    }
+
+    if ((optimizer == "shampoo_momentum") | (optimizer == "shampoo_momentum_diagonal")){
+        [preconL_W1, preconR_W1, momentum_W1, useDiag_W1] = shampoo::init_momentum(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, momentum_b1, useDiag_b1] = shampoo::init_momentum(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, momentum_W2, useDiag_W2] = shampoo::init_momentum(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, momentum_b2, useDiag_b2] = shampoo::init_momentum(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, momentum_W_fc, useDiag_W_fc] = shampoo::init_momentum(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, momentum_b_fc, useDiag_b_fc] = shampoo::init_momentum(b_fc, epsilon, diagThreshold)
+    }
+    if ((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+        [preconL_W1, preconR_W1, stepCounter_W1, bufferL_W1, bufferR_W1, momentum_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1, useDiag_W1] = shampoo::init_heuristic(W1, epsilon, diagThreshold)
+        [preconL_b1, preconR_b1, stepCounter_b1, bufferL_b1, bufferR_b1, momentum_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1, useDiag_b1] = shampoo::init_heuristic(b1, epsilon, diagThreshold)
+        [preconL_W2, preconR_W2, stepCounter_W2, bufferL_W2, bufferR_W2, momentum_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2, useDiag_W2] = shampoo::init_heuristic(W2, epsilon, diagThreshold)
+        [preconL_b2, preconR_b2, stepCounter_b2, bufferL_b2, bufferR_b2, momentum_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2, useDiag_b2] = shampoo::init_heuristic(b2, epsilon, diagThreshold)
+        [preconL_W_fc, preconR_W_fc, stepCounter_W_fc, bufferL_W_fc, bufferR_W_fc, momentum_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc, useDiag_W_fc] = shampoo::init_heuristic(W_fc, epsilon, diagThreshold)
+        [preconL_b_fc, preconR_b_fc, stepCounter_b_fc, bufferL_b_fc, bufferR_b_fc, momentum_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc, useDiag_b_fc] = shampoo::init_heuristic(b_fc, epsilon, diagThreshold)
+    }
+    print(useDiag_W1)
+    print(useDiag_b1)
+    print(useDiag_W2)
+    print(useDiag_b2)
+    print(useDiag_W_fc)
+    print(useDiag_b_fc)
+    data_val_X = X_val
+    data_val_Y = Y_val
+
+    # define the training
+
+    train_losses = matrix(0, rows=epochs, cols=1)
+    train_accuracies = matrix(0, rows=epochs, cols=1)
+    val_accuracies = matrix(0, rows=epochs, cols=1)
+    val_losses = matrix(0, rows=epochs, cols=1)
+    Ntrain = nrow(X_train)
+
+    timestep = 0
+
+    for(epoch in 1:epochs){
+
+        print("Epoch " + epoch + " of " + epochs + " epochs")
+
+        accuracy_value = 0
+        accuracy_count = 0
+        loss_value = 0
+        loss_count = 0
+        
+
+        for(start_index in seq(1, Ntrain, batch_size)){
+            #start_index = (i - 1) * batch_size + 1
+            end_index = min(start_index + batch_size - 1, Ntrain)
+            data_train_X = X_train[start_index:end_index,]
+            data_train_Y = Y_train[start_index:end_index,]
+
+            [softMaxOut, X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred] = modelPredict(data_train_X, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+            predicted_value = rowIndexMax(softMaxOut) - 1
+            accuracy = sum(predicted_value==rowIndexMax(data_train_Y)-1) / length(predicted_value)
+            accuracy_value = accuracy_value + accuracy
+            accuracy_count = accuracy_count + 1
+
+            loss = cross_entropy_loss::forward(softMaxOut, data_train_Y)
+            loss_value = loss_value + loss
+            loss_count = loss_count + 1
+            
+
+            [gradient_lin_layer_W, gradient_lin_layer_b, gradient_W2, rootEvery, preconEveryadient_b2, gradient_W1, gradient_b1] = modelBackward(data_train_Y, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins, X, convOut1, poolOut1, reluOut1, convOut2, poolOut2, reluOut2, pred, softMaxOut)
+
+            if ((optimizer == "shampoo") | (optimizer == "shampoo_diagonal")){
+                [W1, preconL_W1, preconR_W1] = shampoo::update(W1, gradient_W1, lr, preconL_W1, preconR_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1] = shampoo::update(b1, gradient_b1, lr, preconL_b1, preconR_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2] = shampoo::update(W2, gradient_W2, lr, preconL_W2, preconR_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2] = shampoo::update(b2, gradient_b2, lr, preconL_b2, preconR_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc] = shampoo::update(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc] = shampoo::update(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, useDiag_b_fc)
+            }
+            if ((optimizer == "shampoo_momentum") | (optimizer == "shampoo_momentum_diagonal")){
+                [W1, preconL_W1, preconR_W1, momentum_W1] = shampoo::update_momentum(W1, gradient_W1, lr, preconL_W1, preconR_W1, momentum_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1, momentum_b1] = shampoo::update_momentum(b1, gradient_b1, lr, preconL_b1, preconR_b1, momentum_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2, momentum_W2] = shampoo::update_momentum(W2, gradient_W2, lr, preconL_W2, preconR_W2, momentum_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2, momentum_b2] = shampoo::update_momentum(b2, gradient_b2, lr, preconL_b2, preconR_b2, momentum_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc, momentum_W_fc] = shampoo::update_momentum(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, momentum_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc, momentum_b_fc] = shampoo::update_momentum(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, momentum_b_fc, useDiag_b_fc)
+            }
+            if ((optimizer == "shampoo_heuristic") | (optimizer == "shampoo_heuristic_diagonal")){
+                [W1, preconL_W1, preconR_W1, momentum_W1, stepCounter_W1, bufferL_W1, bufferR_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1] = shampoo::update_heuristic(W1, gradient_W1, lr, preconL_W1, preconR_W1, momentum_W1, stepCounter_W1, rootEvery, preconEvery, bufferL_W1, bufferR_W1, preconLInvPowerRoot_W1, preconRInvPowerRoot_W1, useDiag_W1)
+                [b1, preconL_b1, preconR_b1, momentum_b1, stepCounter_b1, bufferL_b1, bufferR_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1] = shampoo::update_heuristic(b1, gradient_b1, lr, preconL_b1, preconR_b1, momentum_b1, stepCounter_b1, rootEvery, preconEvery, bufferL_b1, bufferR_b1, preconLInvPowerRoot_b1, preconRInvPowerRoot_b1, useDiag_b1)
+                [W2, preconL_W2, preconR_W2, momentum_W2, stepCounter_W2, bufferL_W2, bufferR_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2] = shampoo::update_heuristic(W2, gradient_W2, lr, preconL_W2, preconR_W2, momentum_W2, stepCounter_W2, rootEvery, preconEvery, bufferL_W2, bufferR_W2, preconLInvPowerRoot_W2, preconRInvPowerRoot_W2, useDiag_W2)
+                [b2, preconL_b2, preconR_b2, momentum_b2, stepCounter_b2, bufferL_b2, bufferR_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2] = shampoo::update_heuristic(b2, gradient_b2, lr, preconL_b2, preconR_b2, momentum_b2, stepCounter_b2, rootEvery, preconEvery, bufferL_b2, bufferR_b2, preconLInvPowerRoot_b2, preconRInvPowerRoot_b2, useDiag_b2)
+                [W_fc, preconL_W_fc, preconR_W_fc, momentum_W_fc, stepCounter_W_fc, bufferL_W_fc, bufferR_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc] = shampoo::update_heuristic(W_fc, gradient_lin_layer_W, lr, preconL_W_fc, preconR_W_fc, momentum_W_fc, stepCounter_W_fc, rootEvery, preconEvery, bufferL_W_fc, bufferR_W_fc, preconLInvPowerRoot_W_fc, preconRInvPowerRoot_W_fc, useDiag_W_fc)
+                [b_fc, preconL_b_fc, preconR_b_fc, momentum_b_fc, stepCounter_b_fc, bufferL_b_fc, bufferR_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc] = shampoo::update_heuristic(b_fc, gradient_lin_layer_b, lr, preconL_b_fc, preconR_b_fc, momentum_b_fc, stepCounter_b_fc, rootEvery, preconEvery, bufferL_b_fc, bufferR_b_fc, preconLInvPowerRoot_b_fc, preconRInvPowerRoot_b_fc, useDiag_b_fc)
+            }
+        }
+
+        train_losses[epoch,1] = loss_value / loss_count
+        train_accuracies[epoch,1] = accuracy_value/accuracy_count
+
+        [softMaxOut_val, X_val, convOut1_val, poolOut1_val, reluOut1_val, convOut2_val, poolOut2_val, reluOut2_val, pred_val] = modelPredict(X_val, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+
+        predicted_value_val = rowIndexMax(softMaxOut_val) - 1
+        accuracy_val = sum(predicted_value_val==rowIndexMax(Y_val)-1) / length(predicted_value_val)
+
+        val_accuracies[epoch,1] = accuracy_val
+
+        loss = cross_entropy_loss::forward(softMaxOut_val, Y_val)
+        val_losses[epoch,1] = loss
+    }
+
+
+    # define the testing
+
+    [softMaxOut_test, X_test, convOut1_test, poolOut1_test, reluOut1_test, convOut2_test, poolOut2_test, reluOut2_test, pred_test] = modelPredict(X_test, W1, b1, W2, b2, W_fc, b_fc, h_ins, w_ins, c_ins)
+
+    predicted_value = rowIndexMax(softMaxOut_test) - 1
+    accuracy = sum(predicted_value==rowIndexMax(Y_test)-1) / length(predicted_value)
+
+    loss = cross_entropy_loss::forward(softMaxOut_test, Y_test)
+
+    print("Test Accuracy of " + optimizer + " = " + accuracy)
+    print("Test Loss of " + optimizer + " = " + loss)
+
+    if (accuracy > 0.7){
+        print("Test passed")
+    } else{
+        print("Test failed")
+    }
+
+}
+
diff --git a/src/test/scripts/applications/nn/component/shampoo_test.py b/src/test/scripts/applications/nn/component/shampoo_test.py
new file mode 100644
index 00000000000..94028ca443d
--- /dev/null
+++ b/src/test/scripts/applications/nn/component/shampoo_test.py
@@ -0,0 +1,225 @@
+import numpy as np
+
+# update shampoo 
+def update_shampoo(X, dX, lr, preconL, preconR, useDiag):
+    if(not useDiag):
+
+        preconL = preconL + dX @ dX.T
+        preconR = preconR + dX.T @dX
+
+        LEigenvalue, LEigenvector = np.linalg.eig(preconL)
+        preconLInvPowerRoot = LEigenvector @ np.diag(LEigenvalue**(-0.25)) @ LEigenvector.T
+
+        REigenvalue, REigenvector = np.linalg.eig(preconR)
+        preconRInvPowerRoot = REigenvector @ np.diag(REigenvalue**(-0.25)) @ REigenvector.T
+
+        X = X - lr * preconLInvPowerRoot @ dX @ preconRInvPowerRoot
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+    else:
+        n = dX.shape[0]
+        m = dX.shape[1]
+
+        preconL = preconL + (dX**2).sum(axis=1, keepdims=True)          
+        preconR = preconR + (dX**2).sum(axis=0, keepdims=True)
+
+        preconLScale = preconL**(-0.25)        
+        preconRScale = preconR**(-0.25)
+
+        preconLMatrix = preconLScale @ np.ones(shape=[1, m])
+        preconRMatrix = np.ones(shape=(n, 1)) @ preconRScale
+
+        scaledGrad = dX * preconLMatrix
+        scaledGrad = scaledGrad * preconRMatrix
+
+        X = X - lr * scaledGrad
+
+    return(X, preconL, preconR)
+
+# init shampoo
+
+def init_shampoo(X, epsilon, useDiagThreshold):
+    if((X.shape[0] > useDiagThreshold) or (X.shape[1] > useDiagThreshold)):
+        preconL = np.full(shape=(X.shape[0], 1), fill_value=epsilon, dtype=np.float64)
+        preconR = np.full(shape=(1, X.shape[1]), fill_value=epsilon, dtype=np.float64)
+        useDiag = True
+    else:
+        preconL = np.eye(X.shape[0], dtype=np.float64) * epsilon
+        preconR = np.eye(X.shape[1], dtype=np.float64) * epsilon
+
+        useDiag = False
+    return(preconL, preconR, useDiag)
+
+# update shampoo 
+def update_shampoo_momentum(X, dX, lr, preconL, preconR, momentum, useDiag):
+    momentum = 0.9 * momentum + (0.1)*dX
+    if(not useDiag):
+
+        preconL = preconL + dX @ dX.T
+        preconR = preconR + dX.T @dX
+
+        LEigenvalue, LEigenvector = np.linalg.eig(preconL)
+        preconLInvPowerRoot = LEigenvector @ np.diag(LEigenvalue**(-0.25)) @ LEigenvector.T
+
+        REigenvalue, REigenvector = np.linalg.eig(preconR)
+        preconRInvPowerRoot = REigenvector @ np.diag(REigenvalue**(-0.25)) @ REigenvector.T
+
+        X = X - lr * preconLInvPowerRoot @ momentum @ preconRInvPowerRoot
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+    else:
+        n = dX.shape[0]
+        m = dX.shape[1]
+
+        preconL = preconL + (dX**2).sum(axis=1, keepdims=True)          
+        preconR = preconR + (dX**2).sum(axis=0, keepdims=True)
+
+        preconLScale = preconL**(-0.25)        
+        preconRScale = preconR**(-0.25)
+
+        preconLMatrix = preconLScale @ np.ones(shape=[1, m])
+        preconRMatrix = np.ones(shape=(n, 1)) @ preconRScale
+
+        scaledGrad = momentum * preconLMatrix
+        scaledGrad = scaledGrad * preconRMatrix
+
+        X = X - lr * scaledGrad
+
+    return(X, preconL, preconR, momentum)
+
+# init shampoo
+
+def init_shampoo_momentum(X, epsilon, useDiagThreshold):
+    if((X.shape[0] > useDiagThreshold) or (X.shape[1] > useDiagThreshold)):
+        preconL = np.full(shape=(X.shape[0], 1), fill_value=epsilon, dtype=np.float64)
+        preconR = np.full(shape=(1, X.shape[1]), fill_value=epsilon, dtype=np.float64)
+        useDiag = True
+    else:
+        preconL = np.eye(X.shape[0], dtype=np.float64) * epsilon
+        preconR = np.eye(X.shape[1], dtype=np.float64) * epsilon
+
+        useDiag = False
+
+    momentum = X * 0
+    return(preconL, preconR, momentum, useDiag)
+
+# update shampoo 
+def update_shampoo_heuristic(X, dX, lr, preconL, preconR, momentum, stepCounter, rootEvery, preconEvery, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot, useDiag):
+    momentum = 0.9 * momentum + (0.1)*dX
+    if(not useDiag):
+        bufferL = bufferL + (dX @ dX.T)
+        bufferR = bufferR + (dX.T @dX)
+
+        if ((stepCounter > 0) and (stepCounter % preconEvery == 0)):
+            preconL = preconL + bufferL
+            preconR = preconR + bufferR
+            bufferL = bufferL * 0
+            bufferR = bufferR * 0
+
+        if ((stepCounter > 0) and (stepCounter % rootEvery == 0)):
+            LEigenvalue, LEigenvector = np.linalg.eig(preconL)
+            preconLInvPowerRoot = LEigenvector @ np.diag(LEigenvalue**(-0.25)) @ LEigenvector.T
+
+            REigenvalue, REigenvector = np.linalg.eig(preconR)
+            preconRInvPowerRoot = REigenvector @ np.diag(REigenvalue**(-0.25)) @ REigenvector.T
+
+        X = X - lr * preconLInvPowerRoot @ momentum @ preconRInvPowerRoot
+  
+  # Diagonal Shampoo:
+  # Memory-efficient approximation for large parameter matrices
+    else:
+        n = dX.shape[0]
+        m = dX.shape[1]
+
+        bufferL = bufferL + (dX**2).sum(axis=1, keepdims=True)    
+        bufferR = bufferR + (dX**2).sum(axis=0, keepdims=True)
+
+        if ((stepCounter > 0) and (stepCounter % preconEvery == 0)):
+            preconL = preconL +  bufferL     
+            preconR = preconR + bufferR
+            bufferL = bufferL * 0
+            bufferR = bufferR * 0
+        
+        if ((stepCounter > 0) and (stepCounter % rootEvery == 0)):
+            preconLInvPowerRoot = preconL**(-0.25)        
+            preconRInvPowerRoot = preconR**(-0.25)
+
+        preconLMatrix = preconLInvPowerRoot @ np.ones(shape=[1, m])
+        preconRMatrix = np.ones(shape=(n, 1)) @ preconRInvPowerRoot
+
+        scaledGrad = momentum * preconLMatrix
+        scaledGrad = scaledGrad * preconRMatrix
+
+        X = X - lr * scaledGrad
+
+    return(X, preconL, preconR, momentum, stepCounter, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot)
+
+# init shampoo
+
+def init_shampoo_heuristic(X, epsilon, useDiagThreshold):
+    if((X.shape[0] > useDiagThreshold) or (X.shape[1] > useDiagThreshold)):
+        preconL = np.full(shape=(X.shape[0], 1), fill_value=epsilon, dtype=np.float64)
+        preconR = np.full(shape=(1, X.shape[1]), fill_value=epsilon, dtype=np.float64)
+        preconLInvPowerRoot = preconL**(-0.25)
+        preconRInvPowerRoot = preconR**(-0.25)
+        useDiag = True
+    else:
+        preconL = np.eye(X.shape[0], dtype=np.float64) * epsilon
+        preconR = np.eye(X.shape[1], dtype=np.float64) * epsilon
+        preconLInvPowerRoot = np.eye(X.shape[0], dtype=np.float64) * epsilon**(-0.25)
+        preconRInvPowerRoot = np.eye(X.shape[1], dtype=np.float64) * epsilon**(-0.25)
+
+        useDiag = False
+
+    momentum = X * 0
+    bufferR = preconR * 0
+    bufferL = preconL * 0
+    stepCounter = 0
+    momentum = X * 0
+    return(preconL, preconR, stepCounter, bufferL, bufferR, momentum, preconLInvPowerRoot, preconRInvPowerRoot, useDiag)
+
+n = 5
+m = 5
+epsilon = 1e-4
+lr = 0.005
+diagThreshold = 10
+rootEvery=10
+preconEvery=10
+
+# define weight matrix
+X = np.array([
+    [ 0.12, -0.45,  0.33,  0.08, -0.19],
+    [-0.27,  0.41, -0.05,  0.22,  0.14],
+    [ 0.09, -0.31,  0.26, -0.48,  0.37],
+    [ 0.44,  0.06, -0.29,  0.15, -0.11],
+    [-0.38,  0.24,  0.17, -0.07,  0.52],
+], dtype=np.float64)
+
+# define gradient
+dX_main = np.array([
+    [ 0.015, -0.022,  0.008,  0.031, -0.012],
+    [-0.009,  0.027, -0.014,  0.005,  0.019],
+    [ 0.021, -0.006,  0.011, -0.025,  0.004],
+    [-0.018,  0.013, -0.029,  0.007, -0.016],
+    [ 0.010, -0.017,  0.024, -0.003,  0.028],
+], dtype=np.float64)
+
+for diagThreshold in (1, 10):
+    X_py = X.copy()
+    dX = dX_main.copy()
+
+    # preconL_py, preconR_py, useDiag_py = init_shampoo(X_py, epsilon, diagThreshold)
+    # X_py, preconL_py, preconR_py = update_shampoo(X_py, dX, lr, preconL_py, preconR_py, useDiag_py)
+
+    # preconL_py, preconR_py, momentum_py, useDiag_py = init_shampoo_momentum(X_py, epsilon, diagThreshold)
+    # X_py, preconL_py, preconR_py, momentum_py = update_shampoo_momentum(X_py, dX, lr, preconL_py, preconR_py, momentum_py, useDiag_py)
+
+    preconL, preconR, stepCounter, bufferL, bufferR, momentum, preconLInvPowerRoot, preconRInvPowerRoot, useDiag = init_shampoo_heuristic(X_py, epsilon, diagThreshold)
+    X_py, preconL, preconR, momentum, stepCounter, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot = update_shampoo_heuristic(X_py, dX, lr, preconL, preconR, momentum, stepCounter, rootEvery, preconEvery, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot, useDiag)
+    
+    print("diagThreshold: " + str(diagThreshold))
+    print(X_py)
+
+ 
\ No newline at end of file
diff --git a/src/test/scripts/applications/nn/component/shampoo_test2.dml b/src/test/scripts/applications/nn/component/shampoo_test2.dml
new file mode 100644
index 00000000000..750bdeef7c0
--- /dev/null
+++ b/src/test/scripts/applications/nn/component/shampoo_test2.dml
@@ -0,0 +1,34 @@
+source("scripts/nn/optim/shampoo.dml") as shampoo
+
+X_main = matrix("0.12 -0.45  0.33  0.08 -0.19 -0.27  0.41 -0.05  0.22  0.14  0.09 -0.31  0.26 -0.48  0.37 0.44  0.06 -0.29  0.15 -0.11 -0.38  0.24  0.17 -0.07  0.52",
+  rows=5, cols=5
+)
+
+dX_main = matrix("0.015 -0.022  0.008  0.031 -0.012 -0.009  0.027 -0.014  0.005  0.019 0.021 -0.006  0.011 -0.025  0.004 -0.018  0.013 -0.029  0.007 -0.016 0.010 -0.017  0.024 -0.003  0.028",
+  rows=5, cols=5
+)
+
+epsilon = 1e-4
+lr = 0.005
+diagThreshold = 10
+rootEvery = 10
+preconEvery = 10
+
+for (diagThreshold in seq(1, 10, 9)){
+  X = X_main
+  dX = dX_main
+  #[preconL, preconR, useDiag] = shampoo::init(X, epsilon, diagThreshold)
+  #[X, preconL, preconR] = shampoo::update(X, dX, lr, preconL, preconR, useDiag)
+
+  #[preconL, preconR, momentum, useDiag] = shampoo::init_momentum(X, epsilon, diagThreshold)
+  #[X, preconL, preconR, momentum] = shampoo::update_momentum(X, dX, lr, preconL, preconR, momentum, useDiag)
+
+  [preconL, preconR, stepCounter, bufferL, bufferR, momentum, preconLInvPowerRoot, preconRInvPowerRoot, useDiag] = shampoo::init_heuristic(X, epsilon, diagThreshold)
+  [X, preconL, preconR, momentum, stepCounter, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot] = shampoo::update_heuristic(X, dX, lr, preconL, preconR, momentum, stepCounter, rootEvery, preconEvery, bufferL, bufferR, preconLInvPowerRoot, preconRInvPowerRoot, useDiag)
+  
+  print("diagThreshold: " + diagThreshold)
+  print(X)
+}
+
+
+