Created benford_correlation(x) function

[gubertoli]

Hi,

I have been using tsfresh for about 2 months, one attribute that came up to my mind in my cases is to use the deviation from Newcomb-Benford distribution, specifically the correlation between a data array and the expected distribution.

Newcomb-Benford distribution considers a log distribution of the first digits from data.

This analysis has been used for fraud and anomaly detection. More uses - https://scholar.google.com/scholar?q=benford+law+time+series

This is my first pull request to a major package, please if anything is wrong, please let me know.

[coveralls]

Coverage increased (+0.6%) to 97.093% when pulling 1590938 on gubertoli:master into 8dd3f72 on blue-yonder:master.

[nils-braun]

Hi @gubertoli!
Thank you very much for your pull request! We are super happy for new code, especially if they add more feature calculators! The PR is well documented and adds a very reasonable feature calculator.

So, welcome in the open-source community and welcome to tsfresh!

As you said that this is your first commit to a "major" package, I was especially nitpicky :-) So do not be discouraged by the number of review comments.

One thing which is however crucial is that you add a test for your calculator. Please have a look into the file test_feature_calculators.py and add one for your calculator there. Test it for small, large, negative, nan, floating-point, integer, whatever-you-can-think-of numbers. I will not approve the PR without tests ;-)

[nils-braun]

Nice, looks very good now!
Once I understood why the result of the equally distributed numbers is nan and if this is expected, we can merge!

[nils-braun]

Thanks again @gubertoli for your work! Appreciate the changes!

[gubertoli]

Thanks again @gubertoli for your work! Appreciate the changes!

Thank you @nils-braun for the patience and all the processual knowledge shared during this PR.