cgmm

cgmm provides Conditional Gaussian Mixture Models that are fully compatible with scikit-learn. It lets you fit a standard GaussianMixture on your data, then condition on a subset of variables to obtain the posterior distribution of the remaining ones.

Typical applications:

• Multimodal regression (E[y | X=x] and predictive bands)
• Scenario simulation and stochastic forecasting
• Imputation of missing values
• Inverse problems (e.g. kinematics, finance, volatility)

---

Features

• GMMConditioner - take a fitted GaussianMixture, choose conditioning indices, and get a new mixture over the target block.
• ConditionalGMMRegressor - sklearn-style regressor wrapper, exposes .predict and .predict_cov.
• Compatible with scikit-learn pipelines & tools.
• Support for multi-modal posteriors (mixtures, not just means).
• Well-tested, BSD-3 licensed.

---

Installation

pip install cgmm

---

Quick Start

import numpy as np
from sklearn.mixture import GaussianMixture
from cgmm import GMMConditioner, ConditionalGMMRegressor

# Example data: 3 correlated features
rng = np.random.default_rng(0)
X = rng.normal(size=(1000, 3))

# --- Low-level: condition a fitted scikit-learn GaussianMixture ---
gmm = GaussianMixture(n_components=3, covariance_type="full", random_state=0).fit(X)

# Condition on the first coordinate; get a mixture over the remaining two
cond = GMMConditioner(gmm, cond_idx=[0]).precompute()
gmmy = cond.condition([0.5])        # returns a GaussianMixture over y
samples = gmmy.sample(5)[0]         # sample from p(y | X0=0.5)

# --- Regressor: learns its own joint GMM over [features, targets] ---
feats, target = X[:, :1], X[:, 1:]  # predict dims 1,2 from dim 0
reg = ConditionalGMMRegressor(n_components=3, random_state=0).fit(feats, target)
y_mean = reg.predict([[0.5]])       # E[y | X0=0.5]

---

Examples

• 2D conditional demo: scatter + contours + conditional slice.
  https://cgmm.readthedocs.io/en/latest/examples/conditional_2d_gmm.html
• VIX scenario generation: model daily log-VIX changes and simulate stochastic futures.
  https://cgmm.readthedocs.io/en/latest/examples/vix_predictor.html

See the documentation for full tutorials: https://cgmm.readthedocs.io/

---

Documentation

Full documentation is hosted on Read the Docs: https://cgmm.readthedocs.io/

Includes:

• API reference
• Tutorials and examples
• Background on conditional GMMs

---

Contributing

Contributions are welcome! Typical workflow:

git clone https://github.com/your-org/cgmm.git
cd cgmm
poetry install -E docs --with dev
pre-commit install

• Format & lint: make precommit
• Build docs locally: make docs
• Bump version: make bump-patch (or minor/major)
• Push tags → GitHub Actions → PyPI release

Please open issues or pull requests on GitHub.

---

License

BSD-3-Clause License © 2025 Thijs van den Berg