MNIST Training & Testing Automation

View training insights! Visit the live dashboard to see real-time training metrics, model performance, and automated pipeline results.

This repository provides a fully automated pipeline for training and testing a simple neural network on the MNIST dataset using PyTorch. All steps are orchestrated via GitHub Actions, so you get fresh results and logs every time you push or on a schedule.

Features

• Automatic Training & Testing: Runs on every push, manual trigger, or every 6 hours
• Rich Visualizations: Training loss and accuracy graphs with sample prediction images
• Detailed Logs: Beautiful markdown reports with embedded charts and images
• Model Persistence: Trained model weights are saved and version-controlled
• Training Dashboard: View training insights live on GitHub Pages
• Easy to Extend: Simple, clean code structure for experimentation

Project Structure

git-ml/
├── train_mnist.py          # Main training script
├── utils.py               # Utility functions for logging & visualization
├── requirements.txt        # Python dependencies
├── train_output.md         # Training metrics & graphs
├── test_output.md          # Test results & sample predictions
├── mnist_model.pt          # Trained model weights
├── images/                # Generated plots & sample images
├── docs/                   # GitHub Pages website
│   ├── index.html            # Training dashboard page
│   ├── style.css             # Responsive styling
│   └── script.js             # Dashboard functionality
└── .github/workflows/      # Automation workflows
    └── train_mnist.yml       # Training pipeline
    └── pages.yml

Technical Details

Model Architecture

• Input: 28×28 grayscale images (784 features)
• Output: 10 classes (digits 0-9)
• Optimizer: SGD with learning rate 0.01
• Loss Function: Cross-entropy loss

Training Configuration

• Epochs: 15
• Batch Size: 64
• Dataset: MNIST (60k train, 10k test)
• Device: Auto-detection (CUDA if available, else CPU)

Performance Metrics

• Model Type: Simple feedforward neural network
• Dataset: MNIST handwritten digits
• Training Time: ~2-3 minutes on CPU

Quick Start

Option 1: View the Live Dashboard

Visit the training dashboard to view real-time training metrics, model performance insights, and automated pipeline results!

Option 2: Run Locally

1. Clone the repository:

   git clone https://github.com/raselmandol/git-ml.git
   cd git-ml

2. Install dependencies:

   pip install -r requirements.txt

3. Run training:

   python train_mnist.py

4. View results:

   • Check train_output.md for training metrics and graphs
   • Check test_output.md for test results and sample predictions
   • Images are saved in the images/ folder

Option 3: GitHub Actions (Automated)

The workflow automatically runs on:

• Every push to main branch
• Manual trigger via Actions tab
• Every 6 hours (scheduled)

Results are automatically committed back to the repository!

Need detailed setup instructions? Check out the comprehensive setup guide!

Example Output

Training Metrics & Visualizations

The automated pipeline generates comprehensive training reports with interactive visualizations:

Training Progress:

Test Results & Sample Predictions

The model is thoroughly evaluated on the test set, with sample predictions displayed:

Sample MNIST Predictions:

True labels vs Predicted labels are shown for each sample

Customization

Want to experiment? Here are some ideas:

Model Improvements

• Add more hidden layers or change activation functions in train_mnist.py
• Experiment with different optimizers (Adam, AdamW, etc.)
• Try data augmentation techniques
• Implement batch normalization or dropout

Visualization Enhancements

• Modify utils.py to add confusion matrices
• Create loss/accuracy comparison charts
• Add model architecture diagrams

Automation Extensions

• Add model performance benchmarking
• Implement automatic hyperparameter tuning
• Set up notification systems for training completion

Contributing

Contributions are welcome! Here's how you can help:

1. Fork the repository
2. Create a feature branch: git checkout -b feature-name
3. Make changes and add tests
4. Commit changes: git commit -m 'Add feature description'
5. Push to the branch: git push origin feature-name
6. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• PyTorch team for the amazing deep learning framework
• GitHub Actions for seamless CI/CD
• MNIST dataset creators for the classic benchmark
• The open-source community for inspiration and tools
• Copilot, Claude Sonnet 4, GPT-4.1: For always helping me debug errors, suggest code and fixes, and build this repo--including this README.

---

The last commits to this repository were pushed using koelbit-r3 MCP, which is currently under development.

Star this repo if you found it helpful!

Live Demo • Latest Results • Report Bug • Request Feature