Michele Tonutti

@eliottbrion that's a good idea, but off the top of my head I don't have a good idea on how to implement it. I'll give it some thought – if...

Hey, thanks for pointing this out! Indeed my solution seems to only work with TF1. I added a requirements file where a specify the Keras and TF version – I'll...

@MaxBenChrist Sure! Will come soon.

@MaxBenChrist sadly the approach of passing a dictionary with the functions does not work, because the pickling that occurs during the multiprocessing cannot deal with locally defined functions. The alternatives...

@nils-braun well, having a double feature engineering (half of which in pandas, for instance) kinda goes against the speed and scalability advantages of tsfresh! For now, my solution is to...

Ah yes sorry it's on my company's repo: https://github.com/Pacmed/tsfresh here's the diff, ignore the changes not in `feature_calculators`, as I don't think the code in there is used currently https://github.com/blue-yonder/tsfresh/compare/master...Pacmed:master

@nils-braun I've been meaning to try @dbarbier 's solution but I've been very busy! I'll give it a shot hopefully this week and will let you know.

@nils-braun @dbarbier sorry for the massive delay, but I'm happy to confirm that @dbarbier's code for custom functions works indeed. Great stuff! Shall we proceed with a PR?

This didn't use to work (see my comment above https://github.com/blue-yonder/tsfresh/issues/482#issuecomment-465652297), has it been fixed/changed in recent updates?

@nils-braun amazing! That seems to be working perfectly :) Slightly convoluted having to create a dictionary with both function names *and* callables as keys. For a slight more elegant solution...