numpAI is a lightweight implementation of neural networks using only NumPy. It provides a simple interface to build, train, and evaluate custom neural networks using common layer types.
- Modular architecture with a
Networkclass and manyLayersubclasses - Support for common layer types (
FCLayer,ActivationLayer, etc.) - Forward and backward propagation implemented from scratch
- Almost no external dependencies beyond NumPy
- Easy to understand and use
Do NOT load a model file unless you trained it yourself or fully trust its source.
The load method uses Python’s pickle module, which can execute arbitrary code during deserialization. This can pose serious security risks if the file has been tampered with.
For more information, see:
This project is quite the same. But mine is better 😎.
GNU AGPLv3 License. See LICENSE file for details.