Notes, links, ideas on analyzing time series

This is not a replacement for awesome time_series_python

The key lies in finding a similarity matrix. Review.

Options:

• STOMP (python): Computes a matrix profile. Once you have the matrix profile, you can answer 100 questions. Number 2 about repititions is interesting. Multivariate tutorial.
• Dynamic Time Warping (DTW, python): Creates Euclidean distance between points to compute a matrix. Tutorial from scratch. Differentiable DTW.
• Multivariate representation learning: code@tsai, TNC.
• Change point detection: Kats Benchmark
• Periodicity: periodicity, TBATS and SARIMAX, Hyndman's book discussion, see also tsfresh, catch22 estimates of periods.

• Classical time series: Kats, DARTS

• Deep learning:

  • DARTS
  • Gluon-TS
  • tsai
  • pytorch-forecasting
  • flow-forecast
  • Neural Prophet

• Time Series Classification Benchmark and Datasets: https://timeseriesclassification.com/