Rethinking chunking module

predict-idlab / tsflex

Flexible time series feature extraction & processing

https://predict-idlab.github.io/tsflex/

MIT License

405 stars 26 forks source link

Open jonasvdd opened 3 years ago

jonasvdd commented 3 years ago

The chunking submodule is an amazing feature of tsflex, however it still has some flaws.

It assumes:

[ ] that each series-signal is sampled at a (nearly) fixed frequency) -> hence the fs_dict parameter
[ ] has way to many complexity within a single method (hard to read the code and improve)
[x] does not conveniently support the flexibility of only performing time-based chunking

With this issue, I hope to openly discuss how others see this challenge

jonasvdd commented 3 years ago

Some additional notes:

Isn't chunking a variant of a constrained segmenter
As we don't support multi-indexed data, I think group-based chunking support should be in the roadmap

jonasvdd commented 2 years ago