Convolutional Neural Network (CNN)

This project implements a simple Convolutional Neural Network (CNN) with PyTorch, to perform hand write digit recognition. It demonstrates the basic components of CNNs such as convolution, pooling, activation functions, and fully connected layers.

---

Project Structure

.
├── cnn_utils/ # Utility functions for model, data processing, visualization, etc.
├── data/ # training and test dataset
├── visualization/ # Visualization examples (training curves, predictions, etc.)
├── digi_class.py # Main entry point
├── requirements.txt # Python dependencies
├── venv.sh # Virtual environment setup script
└── README.md

---

Features

• Basic CNN layers implemented with PyTorch
• Forward and backward propagation for training
• Simple image classification (e.g., handwritten digits)
• Early stopping and dropout to avoid overfitting.
• Support fine-tuning and resuming training from checkpoints (with model saving)
• Automatic detection of CUDA GPU acceleration with CPU fallback
• Visualization tools for training history curves and prediction
• Modular design for easy extension

---

Installation

Clone the repository and install dependencies:

git clone https://github.com/swangarch/convolutional_neural_network.git
cd convolutional_neural_network

# Create virtual environment and install dependencies
bash venv.sh

source venv/bin/activate

Usage

Usage:

Visualize train data:

python digi_class.py -v <train_data_csv>

Train a model:

python digi_class.py -t <train_data_csv> [weights.pth]

Predict with a model:

python digi_class.py -p <test_data_csv> [weights.pth]

Notes:

• Training data must be a CSV file where the first column is the label, followed by 784 columns representing a 28x28 image.
• Test data must be a CSV file with 784 columns only (no label column).
• If weights are provided during training, the process will perform fine-tuning.
• If weights are provided during prediction, the trained model will be used. Otherwise, an untrained model will be applied (not recommended).
• The program will automatically check for CUDA support; if unavailable, it will fall back to CPU execution.

Visualization

After training:

• The program will display training and validation accuracy curves (so you can visualize model performance over epochs).

After prediction: The program will:

• Save prediction results into a file named predictions.csv.

• Show example images (sampled from the test set) along with their predicted labels.

---

MNIST hand written digit recognition dataset

98% accuracy on Fashion MNIST dataset

Prediciton example:

---

Fashion MNIST classification dataset

88% accuracy on Fashion MNIST dataset

• 0 T-shirt/top
• 1 Trouser
• 2 Pullover
• 3 Dress
• 4 Coat
• 5 Sandal
• 6 Shirt
• 7 Sneaker
• 8 Bag
• 9 Ankle boot

Prediciton example: