Irregular time series and how to whip them
summary
Irregular series get concrete answers: use Lomb-Scargle periodogram (no imputation) to recover periods like RR Lyrae and the 11‑year sunspot cycle, prefer gatspy N log N for speed on big samples, and apply a weighted lasso with a Gaussian kernel to estimate Granger‑style causal links that downweight clustered or sparsely sampled regions. Watch for skewness of gap distributions, noise color, and abrupt failure beyond ~15% missing data; resampling or linear interpolation biases amplitudes and can add white noise.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:47Irregular time series importance
- 5:44Current solutions and flaws
- 9:13Pitfalls of imputing missing data
- 13:46Periodic signals from irregular data
- 16:54Lomb-Scargle periodogram explained
- 19:23Lomb-Scargle advantages
- 21:10Granger causality for irregular series
- 26:25New Granger causality method
- 27:06Summary
- 27:33Q&A
related talks
36mDrift Detection on Irregular Time Series with Multiple Non-Uniform Seasonal Patterns
30mTime Series Analysis via Matrix Estimation
01h 43m(Serious) Time for Time Series
41mReliably forecasting time-series in real-time
01h 18m(Serious) Time for Time Series
45mData Mining in Astronomical Survey Data of Variable Stars with Python
41mApplying Foundational Models for Time Series Anomaly Detection
24mMachine learning and financial time series analysis