Way too many errors with training/test splits here, quite worrying. IMHO part of the problem is that many libraries provide ML methods with little guards against those errors. The responsibility to use them properly lies solely on the practitioner. twitter.com/random_walker/…