orbit

Why we included this project

Orbit wraps several families of Bayesian forecasting models, including exponential smoothing, local-global and damped local trends, and kernel time-based regression, behind a consistent initialize-fit-predict interface. You can switch approaches without rewriting your pipeline, and the probabilistic backends let you fit with MCMC, MAP, or variational inference depending on how much compute you want to spend. The payoff is forecasts that carry calibrated uncertainty, which is what demand, capacity, or revenue planning actually needs. Regression support, backtesting, and diagnostics help validate the models against real data before they go into production. It sits between hand-rolled forecasting code and heavyweight frameworks, and it is a reasonable pick when you want explainable Bayesian models rather than opaque deep learning.