• The module algorithms.py contains the main code for the Extreme Learning Machine and the implementation of iterative algorithms (Nesterov Accelerated Gradient Descent and Gradien Descent). It contains:

  • Classes for the fixed and the parameterized layer
  • A class for the ELM with one hidden layer and L2 regularization, with:
    • Methods to compute the characteristics of the loss function (alpha, beta, k)
    • Method for training via Nesterov's AGD and GD

• The module runners.py contains functions that carry out all the necessary steps to train an ELM with L2 regularization on a dataset. A runner makes the following:

  • Imports a dataset from the UCI repo and prepares the data for training
  • Creates an instance of the ELM
  • Runs training

• The module data_utils.py contains several function to import data from the UCI repository. Many functions are not used in the final implementation. The relevant ones are:

  • import uci_dataset_by_id (imports a dataset from the UCI repository based on its UCI id)
  • encode_dataset (prepares the data for training by performing one-hot encoding on categorical variables)

• The module math_utils.py contains functions to perform mathematical operations useful for training, such as the implementation of activation functions or of strategies for random initialization of weights

• The module linear_algebra_utils.py contains a function that runs the Scipy solver

• The file info_data.csv contains information about the datasets used for evaluation

• The file hyperparameter_evaluation_data.csv contains the results of the evaluation

• The file results_analysis_AGD.txt contains the results of the statistical analysis that have been carried out to extract the insights in the report.