kdexd/digit-classifier

Implements a fully-connected multilayer neural network with sigmoid activation using backpropagation, following a scikit-learn-style API (`fit`, `predict`, `validate`) for clarity and usability. The architecture carefully resolves indexing inconsistencies between MATLAB and NumPy conventions by explicitly separating weights and biases for each layer (excluding the input layer). Suitable for educational purposes and compatible with NumPy's linear algebra operations without external deep learning frameworks.

787 stars. No commits in the last 6 months.