Deep Belief Networks for Classification

This repository demonstrates the utilization of Deep Belief Networks (DBNs) to significantly enhance the accuracy of Multilayer Perceptrons (MLPs) on classification tasks.

Repository Contents:

• rbm.py: Implementation of the Restricted Boltzmann Machine (RBM) class.
• dbn.py: Implementation of the Deep Belief Network (DBN) class.
• dnn.py: Implementation of the Deep Neural Network (DNN) class, comprising a DBN followed by a classification layer.
• vae.py: Implementation of the Variational Autoencoder (VAE) class.
• train.py: Contains training functions:
  • train_RBM: Unsupervised training for RBM.
  • train_DBN: Unsupervised training for DBN.
  • pretrain: Unsupervised training for the DBN within a DNN.
  • backpropagation: Standard training loop for DNN.
  • train_VAE: Standard training loop for VAE.
• test.py: Contains testing functions:
  • test: Computes the accuracy of a model on testing samples.
  • save_images: Saves images in a folder.
  • display_images: Displays images.
  • read_mnist: Loads the MNIST dataset.
  • read_alpha_digit: Loads the BinaryAlphaDigit dataset.

Data Sources:

To use read_mnist and read_alpha_digit you need to download the datasets from THIS CLICKABLE LINK.
The data have to be in a folder named data.

• MNIST Dataset: To load the MNIST dataset, utilize the read_mnist function provided in test.py.
• BinaryAlphaDigit Dataset: To load the BinaryAlphaDigit dataset, use the read_alpha_digit function provided in test.py.

Note: Ensure you have necessary dependencies installed before running the scripts. You will need PyTorch to use the models.