Closed munichpavel closed 4 years ago
Fixed using above SO answer at https://github.com/munichpavel/fake-data-for-learning/commit/b43240bf7b8b0c9df31ea4765587e4c7227c7c97
No need for internal representation or restriction on underscores in values, removed.
I should have known better than to trust an SO solution with 1 vote.
Will just live with sklearn behavior of sorting strings in alphabetical (lexicographical) order.
This has become a bug as I want to sample from a bayesian network conditioned on some values, as I would have to enter the values as a dict with the hacked value names (e.g. "aMale", "bFemale") to override the default sklearn / bumpy ordering