[question] How to get full-ranked categorical data?

[shivamkalra]

My data consists of four columns, lets say a, b, c, d. Where a and b is categorical data with 3 and 5 categories each. My formula is d = C(a) + C(b) + c and even tried d = 0 + C(a) + C(b) + c but none of them is giving me full-ranked data so that I've all 3 and 5 categories in my final input data. Is there way to force patsy to give me full-ranked data?

[njsmith]

There isn't any way baked into patsy currently.

You could do it "by hand" by defining a custom constrast coding scheme -- see https://patsy.readthedocs.io/en/latest/categorical-coding.html

Something like C(a, np.eye(3)) for a variable with 3 levels would get you started. It wouldn't much more work to define a proper coding class that automatically handles different numbers of levels and gives you nicer column names -- search for MyTreat in the docs linked above for an example. And the next step beyond that would be to write up some tests and submit a pull request adding your coding class to patsy itself :-)

(Closing as a duplicate of #60.)