Closed Vayel closed 4 years ago
I think I just gave my "biased" opinion and didn't like explain_predictions
:). I really like explain_influence
though.
I like it as well, but it doesn't tell use what the influence is on. Would you call the methods explain_influence
and explain_performance
?
Maybe it's just me but I find that explain_influence
is enough informative. I think it's pretty obvious that the influence we're hinting at is between the features and the target, which is why users are using the package in the first place. This is Python, not C# :). explain_performance
seems good to me.
What I mean is that explain_performance
is also the explanation of an influence (on the performance). But I'm ok with using explain_influence
for the predictions.
Ah, I see what you mean... I'll have a think about it.
@MaxHalford do you remember why we replaced
explain_predictions()
withexplain_bias()
?In theory, it should be
explain_influence_on_predictions/performance()
but it's a bit long. Maybe we can just useexplain_predictions()
? But since we gave up this name, I guess we had a good reason.