The transformed marker matrix was flatten yielding to uncorrect result due to wrong indexing.
The transformation makes use of pandas dataframes which are also clearer than numpy arrays.