I have an sklearn pipeline consisting of a TfidfVectorizer
and a SGDClassifier(loss='log')
, fitted on a multi-label training dataset. When I then use ELI5 explain_prediction
on a sample (using the vectorizer and classifier from pipeline), I get different probabilities than when I use pipeline.predict_proba
on the same sample. Why is this?
Asked
Active
Viewed 115 times
0
-
1Can you please provide a [minimal, reproducible example](https://stackoverflow.com/help/minimal-reproducible-example) of your code? – Kim Tang Oct 15 '20 at 14:02
-
Have you set `random_state` to the same values? Can you share the relevant part of your eli5 and sklearn model initialisation? – Prayson W. Daniel Oct 16 '20 at 05:56