Open houssamzenati opened 2 weeks ago
Yes, we need homogeneity !
I also think that n_discarded is an internal parameter that should not be returned (but could be logged in the future)
n_discarded is not internal, it's important to be aware if trim argument is too high. @houssamzenati suggested a warning instead, and in that case we could remove it from the returned values. WDYT @bthirion?
But I'm afraid there are 10 internal parameters of that kind and that we're focusing on one particular ? Bzw, I have the feeling that we should never use trimming.
Looking at the documentation, it would be nice to have unified names on what estimators return
example for
mediation_IPW
example for
mediation_dml
:moreover, is it really useful for the user of the library to have
n_discarded
?? should we discard it for all estimators??