Here's one for the data scientists and ML Engineers.

Someone set a literal date feature (not month, not season, but date) as a categorical feature... as a string type 🥺

I don't trust this model will perform for long

  • 2
    Actually, it might not be completely stupid if there's not meaning in a strict sense of ordering, as in, having recentness be a feature.

    If you just want to find a correlation *without* considering temporal ordering as an implicit feature, it might even make sense.
  • 1
    I mean, it's the reason why you one hot encode categorical features instead of just assigning an int value enum-style, to prevent wrong correlation being derived by arbitrary ordering of the enum.
Add Comment