numpy-ml

Why we included this project

numpy-ml implements a wide range of machine learning algorithms from scratch, using nothing but NumPy. The code is written to be read rather than to win benchmarks, which makes it a good place to see how a hidden Markov model, a Gaussian mixture, or a transformer-style attention block actually works. Students and engineers brushing up on fundamentals will find it easier to follow than optimized library internals or dense research papers, and it also works as a reference implementation to adapt when prototyping. Coverage runs from classical models like LDA and n-gram smoothing to modern components such as LSTMs, ResNet blocks, and WaveNet-style convolutions.