saitej123/ERA-V1--PyTorch

Handwritten Digit Recognition for MNIST Dataset with PyTorch

This repository provides use PyTorch to train MNIST dataset

Table of Contents

• Introduction
• Features
• Requirements
• Files Details
• How To Use
• Results

Introduction

This repository contains Python scripts for training and testing a Convolutional Neural Network (CNN) model on the MNIST dataset for recognizing handwritten digits.

Features

• Leveraging PyTorch for efficient computational execution
• Preprocessing the MNIST dataset utilizing data augmentation techniques
• Visualizing the processed images and their respective labels
• Training a Convolutional Neural Network model on the processed data
• Evaluating the trained model on the test portion of the data
• Plotting the accuracy and loss curves during training and testing

Requirements

• Python 3.10
• PyTorch 2.0
• torchvision
• Matplotlib
• CUDA (recommended for GPU)

Files Details

• utils.py: This file contains all the helper functions required for the workflow.

  • get_correct_pred_count: Calculates the number of correct predictions
  • train: Trains the model on the training dataset
  • test: Evaluates the model on the test dataset
  • loss_plots: Plotting the loss , accuracies for training and test

• model.py: This file defines the architecture of the Convolutional Neural Network model Net(nn.Module) used for the task. It includes convolutional layers and fully connected layers.

• S5.ipynb : The main Jupyter notebook that manages the entire process. It imports functions from utils.py and the model from model.py to train and evaluate the model on the MNIST dataset. The notebook orchestrates the overall workflow.

How To Use

• Run the S5.ipynb notebook in a Jupyter notebook environment. If CUDA is available on your machine, the code will automatically use it to speed up computations.

• The notebook includes the following steps:

  1. Data Transformations: Application of transformations on the MNIST dataset for data augmentation and normalization.
  2. Model training: Training the CNN model using the preprocessed training set.
  3. Model testing: Evaluation of the trained model on the test dataset and print of the loss and accuracy.
  4. Performance visualization: Plotting of accuracy and loss graphs for both the training and test datasets.

Results

The final trained model achieves an accuracy of 99.50% on the MNIST test set.

Model Summary
----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1           [-1, 32, 26, 26]             320
            Conv2d-2           [-1, 64, 24, 24]          18,496
            Conv2d-3          [-1, 128, 10, 10]          73,856
            Conv2d-4            [-1, 256, 8, 8]         295,168
            Linear-5                   [-1, 50]         204,850
            Linear-6                   [-1, 10]             510
================================================================
Total params: 593,200
Trainable params: 593,200
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 0.67
Params size (MB): 2.26
Estimated Total Size (MB): 2.94
----------------------------------------------------------------

Developed by Sai teja (macharlasaiteja@gmail.com)