No Selected Features about tsfresh

[MaxBenChrist]

Can you provide an examplary notebook with data? Otherwise we can not investigate properly.

Also, can you provide a selection report? You get it by setting the write_selection_reportflag of the FeatureSelectionObject to True

[javierorozco]

This ZIP contains demo data and Jupiter Notebook
javier_demo.zip

[MaxBenChrist]

You have too many unimportant features.

from tsfresh.feature_selection import FeatureSignificanceTestsSettings
sett = FeatureSignificanceTestsSettings()
sett.fdr_level = 0.5 # increases the risk of unimportant features being extracted
sett.write_selection_report = True # will create a report with more information

# i only inspect first columns to 50th feature to reduce number of irrelevant features 
X_filtered3 = select_features(X.iloc[:,0:50], y, feature_selection_settings=sett)
X_filtered3.head()

This will extract a__time_reversal_asymmetry_statistic__lag_2, a__mean_change and a__number_peaks__n_1

However, the handling of constant features is internally not done in the best fashion. TSFRESH could extract more features, I am working on that.

[MaxBenChrist]

I looked into that in detail javier. I found some bugs but your problem stays the same. Your samples does not contain enough evidence to make a feature "relevant".

Either increase the FDR level or use another feature selection method

[javierorozco]

Thanks Max for the update, have you pushed the bugs fixing to master?

[MaxBenChrist]

Yes it is commit 7a21b5e

you can install master of tsfresh by pip install git+https://github.com/blue-yonder/tsfresh