/PyTorch-Neural-Network-Implementation

Primary LanguageJupyter Notebook

PyTorch-Neural-Network-Implementation

This project focuses on implementing and comparing various aspects of deep learning, from data loading to training neural networks, using both PyTorch's built-in functionalities and custom implementations. The tasks include calculating execution time, comparing custom and PyTorch data loaders, implementing and training neural networks, and developing a custom back-propagation algorithm.

Project Structure

  • data/: Contains the MNIST dataset and any related files.
  • models/: Contains the implementations of neural network architectures.
  • notebooks/: Jupyter notebooks for experimentation and visualization.
  • scripts/: Python scripts for data loading, training models, and plotting results.
  • results/: Directory to save model checkpoints, loss logs, and performance graphs.
  • README.md: Project documentation.

Requirements

  • Python 3.8+
  • PyTorch
  • torchvision
  • numpy
  • scikit-learn
  • matplotlib
  • seaborn

Tasks

Task 1: Custom Data Loader vs. PyTorch Data Loader

Downloading MNIST Dataset in Google Colab

import torch
from torchvision import datasets, transforms

# Download MNIST dataset
transform = transforms.Compose([transforms.ToTensor()])
train_dataset = datasets.MNIST(root='./data', train=True, transform=transform, download=True)
test_dataset = datasets.MNIST(root='./data', train=False, transform=transform, download=True)