Uncertainty estimation at scale with functime
summary
functime claims to produce prediction intervals for thousands of series locally by combining Polars-powered feature extraction and cross‑validation that runs 1–2 orders of magnitude faster with a conformal method named EnBPI conformal prediction that provides proven marginal and (under extra assumptions) conditional coverage. implementation uses a block bootstrap ensemble to estimate nonconformity and per‑point quantiles, needs roughly 500–10k calibration points in practice and assumes residuals are approximately iid; plotting support for diagnostics is still a work in progress.
AI-written summary · about
chapters (8)
- 0:00Uncertainty estimation intro
- 3:56Conformal prediction importance
- 9:57Traditional uncertainty limits
- 13:50Data exchangeability in conformal
- 19:14Conformal prediction in practice
- 25:36Conformal for time series
- 32:58functime implementation & scale
- 45:01Q&A
related talks
31mfunctime: a next generation ML forecasting library powered by Polars
37mQuantifying Uncertainty in Time Series Forecasting
28mHow I used Polars to build built functime, a next gen ML forecasting library
30mConformal Prediction: uncertainty quantification to humanise models
31mUncertainty quantification: How much can you trust your machine learning model?
32mLeveraging Conformal Prediction for Calibrated Probabilistic Time Series Forecasts
36mQuantifying uncertainty in Machine Learning predictions
22mConformal Inference with Tidymodels