EPFL Deep Learning Classification Project

This repository contains an implementation of various machine learning methods for a classification task, including PCA for dimensionality reduction and multiple deep learning architectures such as a Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and a Vision Transformer (ViT). This project is part of an academic assignment at EPFL.

Table of Contents

• Overview
• Features
• Project Structure
• Requirements
• Installation
• Usage
• Testing
• Acknowledgments

Overview

The goal of this project is to implement and evaluate different machine learning methods for image classification. The repository includes:

• Dummy Classifier: A baseline method that returns random predictions.
• PCA: A principal component analysis implementation to reduce feature dimensionality.
• Neural Networks:
  • MLP (Multi-Layer Perceptron): A fully-connected network for classification.
  • CNN (Convolutional Neural Network): A network that uses convolutional layers for feature extraction.
  • Vision Transformer (ViT): A transformer-based network for image classification.

The project uses TensorBoard to log training metrics and supports training on CPU or GPU.

Features

• Data Processing: Normalization and optional PCA-based dimensionality reduction.
• Modular Design: Multiple architectures (MLP, CNN, ViT) for classification.
• Training & Evaluation: Training routines using PyTorch with support for validation splits.
• Logging: Integration with TensorBoard for monitoring training progress.
• Unit Testing: Tests to validate the implementations.

Requirements

• Python 3.6+
• NumPy
• PyTorch
• TorchInfo
• SciPy
• TensorBoard

Installation

1. Clone the Repository:

   git clone https://github.com/yourusername/EPFL-Deep-Learning-Classification-Project.git
   cd EPFL-Deep-Learning-Classification-Project
   

Optional

python -m venv venv

source venv/bin/activate  # On Windows: venv\Scripts\activate

pip install -r requirements.txt

Usage

The main training and evaluation script is main.py. You can specify various command-line arguments:

• --data: Path to the dataset directory.
• --nn_type: Type of neural network to use (mlp, cnn, or transformer).
• --nn_batch_size: Batch size for training.
• --use_pca: Flag to enable PCA dimensionality reduction.
• --pca_d: Number of principal components if PCA is used.
• --lr: Learning rate for training.
• --max_iters: Number of training epochs.
• --test: Flag to indicate testing on the entire dataset.

Example

python main.py --data dataset --nn_type cnn --nn_batch_size 64 --lr 1e-5 --max_iters 100 --use_pca --pca_d 100

Testing

To run the unit tests, execute:

python test_ms2.py

Acknowledgments

This project was developed as part of the coursework for the EPFL Introduction to Machine Learning class CS-233. Special thanks to the course instructors and TAs.