During the feature extraction process, how could I set the param from a subset of another columns.

[lk1983823]

For example, I want to use "number_crossing_m" to a column named "x1". So I have to set a parameter "m" to it. However, for the param "m", I want to let it be the mean of another column like "x2" or the "x1" itself. How should I do it ?

Thanks.

[kohlrabi90]

Hi @lk1983823 ,
with the default feature extraction function, it is not possible to do this. However, you could run the extraction on each of your time series (columns) separately: first, extract the mean, then adjust the settings, and finally call feature extraction only on one series. Then repeat for the next.
You can use the feature extraction settings to customize "m".

Here is an example how you can set "m"

from tsfresh.feature_extraction import ComprehensiveFCParameters
from tsfresh.feature_extraction import extract_features

# Adjust the value for m
settings = ComprehensiveFCParameters()
settings["number_crossing_m"] = [{'m': your_mean_value}]

results = extract_features(df,  default_fc_parameters=settings)

Additionally, (but more complex), you could write your own custom feature calculator, which does (1) calculate the mean and (2) then call the original feature calculator. However, that only works for a single column (not when you want to use the mean for x2 for calculating x1). I would recommend the first option.