Show HN: Farseer, rust rewrite of prophet with some extra goodies (ryanbieber.github.io)

Farseer is a Rust-based reimplementation of Facebook Prophet that preserves Prophet’s familiar scikit-learn-like API while delivering substantially faster, production-friendly forecasting. It uses a Bayesian model with uncertainty intervals, automatic handling of missing data, outliers and changepoints, and automatic multithreading to use all CPU cores — authors report 5–10× speedups versus Prophet. It supports both Polars (recommended for max speed) and Pandas DataFrames, and ships via pip or as a maturin-built development install. Technically, Farseer adds native observation weights, automatic regressor type detection and standardization, conditional seasonality (apply seasonal patterns only under specified conditions), independent per-holiday priors, and logistic growth with both cap and floor. Models can be saved/loaded as JSON for reproducible deployment and the API remains nearly identical to Prophet’s, lowering migration friction. For the AI/ML community this means much faster iterative model fitting, easier weighting of recent/reliable data, and simpler productionization without heavy Stan dependencies — making robust probabilistic time-series forecasting more scalable and easier to integrate into real-world pipelines.