bab2min/tomotopy

Implements 14+ topic modeling algorithms (LDA, HDP, DTM, CTM, and variants) using Collapsed Gibbs Sampling with SIMD vectorization (AVX512/AVX2/SSE2) for CPU-accelerated inference. Achieves 5-10x faster iteration speeds than Variational Bayes approaches by leveraging modern CPU instruction sets and multicore parallelization. Supports both training document inference and out-of-sample prediction, with model persistence and an interactive web-based visualization viewer.

592 stars.