Calculating timeseries features using get_meta_data() with different periodicity.

[robertcurler]

Hi there,

I am trying to use the Kats.metalearner package to generate a classifier on monthly data. However the standard periodicity is set to 7 in the feature generation call: TsFeatures( window_size: int = 20, spectral_freq: int = 1, stl_period: int = 7,...

So basicly I am initializing a GetMetaData class as follows:

GetMetaData(data=series, all_models=all_models,
            all_params=all_params,
            error_method='rmse',
            min_length=24)#,
            #"stl_period": 12)

And then I follow up with a get_meta_data() call like so: metadata_dict[timeseriesID] = metadata_dict_class[timeseriesID].get_meta_data() Which then runs TsFeatures().transform(self.data) in its first line.

In my limited knowledge of OOP, I tried to funnel the alternative stl_period in the GetMetaData init. call. My approach didn't work.

Is there a way of correctly funneling this alternative setting, or should i skip the get_meta_data() and adjust the settings of the calls present in get_meta_data() manually?

Regards, Robert

[neelabalan]

I think you can call the tune_executor and TsFeatures().transform (with custom args) separately and merge the dictionaries to get the metadata.

[peiyizhang]

Hi Robert, thanks so much for your questions.

We just added this function into GetMetaData module. Now if you would like to apply that, you could try:

GMD = GetMetaData(data=TSdata)
_ = GMD.get_meta_data(stl_period=your_alternative_number)

And this will run TsFeatures(stl_period=your_alternative_number).transform(self.data) in its first line. You could also use alternative window_size, spectral_freq, nbins, lag_size, etc.

Note that you have to install Kats from source again to be able to use this change.

Please let us know if you have any further questions!

[iamxiaodong]

issue resolved, close this issue

[robertcurler]

Hey all,

I finally got around to trying it myself. Works as expected ;)

Thanks a bunch!