Orbit: Probabilistic Forecast with Exponential Smoothing
Time series forecast is an active research topic in academia as well as industry. Although we see an increasing amount of adoptions of machine learning methods in solving some of those forecasting challenges, statistical methods remain powerful while dealing with low granularity data. The Orbit python package implements a family of refined Bayesian exponential smoothing models using the probabilistic sampling languages, Stan and Pyro. Specifically, our proposed models are Seasonal Local and Global Trend (SLGT) and Damped Local Trend (DLT). The model refinements include additional global trend (linear, log-linear, or logistic), transformation for multiplicative form, noise distribution and choice of priors. Orbit also comes with powerful utilities to perform the backtesting tasks. The present notebook will illustrate typical use of Orbit and demonstrate a benchmark study on a rich set of time-series data sets for our proposed models along with other well-known time series models including Prophet and SARIMA.
Edwin Ng is a senior data scientist at Uber where his team builds forecast models to support strategic decisions and optimization for marketing. Prior to Uber, he was building cutting-edge statistical and predictive models in PayPal and Ten-X Commercial. Edwin received a MFE from UCLA Anderson School of Management.