Cyclical/Seasonality Feature Generation

[dreyco676]

I'm wondering if it would be in scope for TSFRESH to have some feature generators that build out features from the DatetimeIndex for use in forecasting applications to capture cyclical events. Some rational can be found here.

New Features:

• Year
• Month
• Day of Year
• Week of Year
• Weekday/Weekend
• Hour of Day
• Minute of Hour
• Second of Min
• etc

TSFRESH could leverage many of the methods available in Pandas DatetimeIndex to automate features efficiently by inferring the frequency and total span of time the dataset covers to only create applicable features.

Further TSFRESH could incorporate SIN and COS transformations for datasets that are continuous.

If these are in scope for this project, I'd be happy to take a stab at a PR. If not, no worries.

[MaxBenChrist]

Sorry for the late answer @dreyco676 . The links that you send me are interesting.

I am not sure how to implement this properly and I am lacking at the application at the moment where I would need something like this. Can you provide an application where those features are useful?

Please keep in mind that tsfresh maps time series to a float, the sin/cos transformation that you proposed map a time series into another time series (or even two!)

[wgova]

@dreyco676 and @MaxBenChrist - have you made any progress on this? Time series decomposition for multiple time series would help in grouping time series discrimination, I think. Please advise me on any progress or useful material