Daniel Bartling

@fkiraly * could you kindly explain why there are changes in the X/y pickle files and what these are? => Frequency got lost in pandas 1.5.0 * reducer changes look...

Hi the issue here is that you use kurtosis and skewness over a window for which your forecasting model(linear regression) creates all constant values. Kurtosis and skewness are not defined...

@fkiraly No kurtosis and skewness are not defined for constant values (since you divide by deviation to the power of 4). And since the forecast generates constant values, you will...

Hi I checked the example by @ngupta23 and the issue seems to be that the way you try to set indices ignoring the instancing of the multiindex: ``` y_global.set_index(["time_series", "date"],...

@fkiraly temporal_train_test_split does not lose the freq, it actually was never there because the way it is assigned (see the full notebook, not the just the script) by ngupta meant...

You are absolutely right @ngupta23 , thanks for the suggestion. Will include that

@ngupta23 @fkiraly Set up the following pull request to raise a warning and also provide a convience function to set freq for multiindex. The solution I presented with convert datetimeindex...