SeanTaylorBenLetham
summary
Facebook builds Prophet, a time‑series tool implemented in R and Python on Stan MCMC, that models real business series with piecewise linear trend and Fourier seasonality plus holiday dummies and Laplace priors on changepoints. Default MAP fits run under a second for exploration, sampling gives full posteriors, and it handles missing data and multiple seasonalities. Adopted for capacity planning, goal baselines and bug spikes, uncertainty calibration varies and users can tweak priors and error structure.
AI-written summary · about
chapters (16)
- 0:00Intro
- 1:03Problems
- 2:26Demand for Business Forecasts
- 6:14Solution
- 11:54Facebook Events
- 15:10A decomposable
- 17:06Stan Code
- 17:47Prophet Forecast
- 18:20Forecast Components
- 19:17Estimating Uncertainty
- 23:26Knobs and Levers
- 26:33How we use Stan
- 28:10Supporting Python and R
- 29:23Democratizing Statistical Tools
- 33:06Takeaways
- 34:49Summary
related talks
28mBuild Facebook's Prophet in PyMC3
41mReliably forecasting time-series in real-time
41mNeural Prophet – A powerful AI framework for Time Series Models
01h 27mBayesian Time Series
37mTracking the tracker: Time Series Analysis in Python From First Principles
40mMulti-Series Forecasting at Scale with StatsForecast
01h 26mApplying Statistical Modeling & Machine Learning to Perform Time-Series Forecasting
34mFast, beautiful and easy Bayesian modelling: Can you have it all?