Better transformation for bounded distributions #3058
Conversation
close #3010 also some PEP8 clean ups and better parameter name for truncated_normal
| @@ -488,32 +499,33 @@ class TruncatedNormal(Continuous): | |||
| Mean. | |||
| sd : float | |||
| Standard deviation (sd > 0). | |||
| a : float (optional) | |||
| lower : float (optional) | |||
There was a problem hiding this comment.
I agree that lower and upper are nicer names, but it is different from scipy.stats.truncnormal.
There was a problem hiding this comment.
we already use mu and sd instead of loc and scale, so I guess two more doesnt hurt ¯\_(ツ)_/¯
Also, since we are calling sd instead of sigma, pretty sure we lose the moral high ground to criticize naming in other packages...
There was a problem hiding this comment.
We should really fix the sd - sigma thing (my fault).
|
@aseyboldt also commented on lower and upper not being bounded which causes some sharp cliffs in the posterior. |
|
Not sure I understand what that means... |
|
@junpenglao Sorry, that was a comment on the original TruncatedNormal implementation. This PR is exactly what I thought was missing :-) |
close #3010
also some PEP8 clean ups
and better parameter name for truncated_normal