This repository contains a lightweight neural network implementation in C. It provides a framework for creating, training, and evaluating neural networks with configurable architectures through JSON configuration.
- Neural network implementation in pure C
- JSON-based configuration for network architecture
- AdamW optimizer
- CSV output for loss and accuracy tracking and results visualization
Use make to build from source and run the executable. The JSON configuration file can be provided as an argument, otherwise the file ./config.json will be used.
- Many hardcoded things, including the input format, output format and metrics
- Minimal error handling
- Only supports feed-forward network
- CPU only