implicit

Why we included this project

For anyone building recommendations from behavioral data rather than explicit ratings, implicit is a solid starting point. It implements the standard collaborative filtering algorithms, including alternating least squares, Bayesian personalized ranking, and logistic matrix factorization, all trained on the signals you already collect, like page views, plays, and purchases, so users never have to rate anything. Training is fast thanks to multithreaded Cython code, and the ALS and BPR models include CUDA kernels, so they can run on GPUs when your dataset gets large. The fit-and-recommend API over sparse user-item matrices makes it easy to drop into an existing pipeline and compare models on your own metrics. It also works well as a reference for learning how the math.