ERA - MNIST Classification using Pytorch

This project demonstrates image classification on the MNIST dataset using PyTorch deep learning library. The MNIST dataset consists of handwritten digit images and their corresponding labels.

Requirements

• Python (>=3.6)
• PyTorch (>=1.7)
• torchvision
• matplotlib
• tqdm

Install the Required Packages on your Environment

For CUDA Enabled Devices:

conda install pytorch torchvision pytorch-cuda=11.8 -c pytorch -c nvidia
conda install matplotlib tqdm

For CPU Only Devices:

conda install pytorch torchvision cpuonly -c pytorch
conda install matplotlib tqdm

Install Jupyter Notebook (if not already installed):

conda install jupyter

Project Structure

The project has the following structure:

1. S5.ipynb : This is the main notebook where you execute the code. It imports functions and classes from other files and contains the main logic for training and testing the model.

   • train_transforms : A data transformation pipeline for the training data.
   • test_transforms : A data transformation pipeline for the test data.

2. utils.py : This file contains utility functions and data loading functions that are commonly used across different modules. It provides the following functionalities:

   • train: A function for training the model.
   • test : A function for testing the model.

3. model.py : This file contains the definitions of the neural network models. It provides the following classes:

   • Net: Neural network model with a specific architecture.

Usage

1. Open the S5.ipynb file in Jupyter Notebook or any compatible environment.

2. Execute the code cells in the notebook to train the model, evaluate it on the test set, and visualize the results.

Code Explanation

The provided code trains a neural network model on the MNIST dataset using PyTorch. It performs the following steps:

1. Imports necessary libraries and modules, including PyTorch, torch.optim, utils, and model.

2. Checking CUDA Availability :

   • The code checks if CUDA is available for GPU acceleration by calling torch.cuda.is_available() . The result is stored in the cuda variable.

3. Data Loading and Transformation:

   • The code defines the data transformations for the training and test datasets using train_transforms and test_transforms.
   • The MNIST dataset is loaded using these transformations, and data loaders are created for both the training and test datasets.

4. Model Definition:

   • The code defines the neural network model architecture using the Net class from the model module.
   • The model is moved to the available device (GPU if CUDA is available, otherwise CPU).
   • A summary of the model architecture is printed using the summary function from the torchsummary module.

5. Training Setup:

   • The code sets up the optimizer (Adam optimizer), learning rate scheduler (step scheduler), loss criterion (F.nll_loss), and the number of epochs for training.

6. Training Loop:

   • The code initiates the training loop for the specified number of epochs.
   • In each epoch, the train function from the utils module is called to train the model on the training data.
   • After training, the test function from the utils module is called to evaluate the model on the test data.
   • The learning rate scheduler is then adjusted accordingly.