MathCAS: Modular Neural Network Training Framework

A minimal, modular PyTorch-based framework for training neural networks on tabular data, with YAML-configurable architecture, logging, and experiment management.

Note: This project is a work in progress. Features and documentation may change as development continues.

Features

• Modular architecture and data loading
• YAML-based experiment configuration
• Logging to console and file (configurable)
• Early stopping, reproducibility, and output management

Quickstart

1. Install dependencies:

   pip install -r requirements.txt
   

2. Prepare your dataset:
   Place your CSV in datasets/ and update templates/regression.yaml as needed.

3. Run training:

   python main.py
   

4. Check outputs:
   Models, logs, and metrics are saved in the outputs/ directory.

Configuration

Edit templates/regression.yaml or create custom templates to change model, training, or data settings.

License

MIT License. See LICENSE.