Time Series Decomposition with PyMC
summary
Interpretable time‑series decomposition built with PyMC separates trend, seasonality and noise using engineered features and actually fits: a linear slope around 0.06 s per lap, pit stop cost ≈21–22 s, and a first‑lap grid penalty ≈0.7–1 s; medium tyres show larger degradation than hard. Seasonality uses many radial basis functions (initially wide + narrow then pruned) and correlated series use hierarchical priors so drivers and teams learn from each other, producing calibrated posteriors and lower residual variance.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:09Formula 1 meets Bayesian modeling
- 5:01Agenda and overview
- 7:51Modeling lap times
- 10:39Tire compound effects
- 12:51Modeling outliers like pit stops and starting positions
- 14:18Seasonality with radial basis
- 18:30Multi-level driver/team models
- 23:05F1 applications
- 27:58Key takeaways
- 29:43Q&A
related talks
01h 27mBayesian Time Series
44mActually using GPs in practice with PyMC
41mReliably forecasting time-series in real-time
01h 27mProbabilistic Python: An Introduction to Bayesian Modeling with PyMC
27mPredicting F1 Race Strategies using ML.NET
01h 43m(Serious) Time for Time Series
01h 26mProbabilistic Programming and Bayesian Computing with PyMC
01h 30mTime Series Analysis with StatsModels