Closed tirthasheshpatel closed 2 years ago
OK. You mention not deprecating the loguniform
and reciprocal
distributions. Would you like to include those changes here?
Would you like to include those changes here?
I have my final project presentation tomorrow so I won't be able to get back to this until the day after tomorrow. Meanwhile, feel free to make the changes and push this change to your main branch if you agree that we should not deprecate loguniform
and reciprocal
distributions. Thanks!
if you agree that we should not deprecate
Haha I don't think we should have to create an entirely new distribution to offer an alternate parameterization, and I don't think distributions should have to use the loc
/scale
parameterization if it's not natural. I guess this just serves as a reminder of why gh-15928 is important. So sure, let's leave loguniform
and reciprocal
without deprecation.
This can wait a little longer. I just want to get it into 1.9.0, so let me know if you'd have trouble getting to it in time for branching.
Best wishes on your presentation @tirthasheshpatel!
Best wishes on your presentation @tirthasheshpatel!
Thanks @mdhaber :smiley:
Thanks for all the changes @mdhaber!
Reference issue
scipy#15889
What does this implement/fix?
Raise a
UserWarning
when the fit method of theloguniform
andreciprocal
distributions is called. Instead of deprecating theloguniform
andreciprocal
distributions, I think moving users away from the[loguniform|reciprocal].fit
method is more user friendly.loguniform
andreciprocal
still provide useful functionality and canonical parametrization for new users or users who don't use thefit
method.