Adopting Open-Source Tools for Time Series Forecasting
summary
A decision framework evaluates actual capabilities not topics: it scores libraries on Data Understanding, Data Preparation, and Modeling & Backtesting and applies them to an energy management case (PJM hourly load). SKTime provides built-in EDA plots, imputation, detrending and detailed sliding/expanding-window evaluation but runs slower; SKForecast delegates preprocessing to pandas, offers a compact fast backtesting API with configurable window splitters, and returns concise fold metrics. The work benchmarks these tools against an internal Schneider Electric pipeline and surfaces concrete tradeoffs for adoption.
AI-written summary · about
chapters (19)
- 0:00Introduction
- 1:45Talk outline
- 3:35Time series foundations
- 5:41Forecasting framework
- 7:25Six phases of forecasting
- 9:53Open-source advantages
- 11:40Tool evaluation criteria
- 13:03Energy usage example
- 14:06Data understanding with SKTime
- 16:59Handling data quality
- 18:01Missing value imputation
- 18:43Outlier detection
- 19:27Trend removal and split
- 20:11Data prep with SKForecast
- 21:38Back testing basics
- 23:05Back testing in SKTime
- 25:36Comparing results
- 27:46Key takeaways
- 29:09Q&A
related talks
30msktime: python toolbox for time series- next-generation AI
25mThe Value of Open Source Short-term Energy Forecasting Demo...
26mTime Series Analysis in Python: Easy Tools for Scientific Insight
01h 26mSktime—Python Toolbox for Time Series: How to Implement Your Own Estimator
44mMethods and Tools for Time Series Data Science Problems
27mBenchmarking Time Series Foundation Models with sktime
40mMulti-Series Forecasting at Scale with StatsForecast
40mDeep Learning and Time Series Forecasting for Smarter Energy