sktime - python toolbox for time series: advanced forecasting - probabilistic, glob...
summary
sktime offers a scikit‑learn–compatible forecasting API with four concrete probabilistic outputs: predict_interval, predict_quantiles, predict_var, and predict_proba (returns TensorFlow distributions). it vectorizes over panel and hierarchical pandas multi‑index data, runs thousands of independent fits (examples use ~43,000 series from M5), and provides an aggregator plus reconciler (OLS/WLS) to enforce coherence; global forecasting uses a windows summarizer to build lag/rolling features for tree models. notebooks run via pip until the conda package updates.
AI-written summary · about
chapters (13)
- 0:00Intro
- 2:08About sktime
- 5:27Philosophy behind sktime
- 9:35Time series
- 14:17Learning tasks
- 18:15probabilistic forecasting
- 27:54frequent points of confusion
- 36:56metrics
- 38:46evaluation interface
- 41:29summary
- 45:30Hierarchical time series
- 46:49Hierarchical and panel data types
- 54:08Hierarchical reconciliation
related talks
26mTime Series Analysis in Python: Easy Tools for Scientific Insight
01h 29msktime: time series anomaly detection, changepoint detection, segmentation
27mBenchmarking Time Series Foundation Models with sktime
39mForecasting With Classical and Machine Learning Methods
27mTime series feature extraction with tsfresh - “get rich or die..
01h 02mDeep learning for time series forecasting and classification in practice
42mFlying Python at 40,000 feet
01h 30mTime Series Analysis with StatsModels