Closed bgroenks96 closed 11 months ago
I believe I did this because the previous version was using the TruncatedBijector
and so I didn't want to touch in the PR where I added has_constant_bijector
, etc.
The downside with the map
approach is that it will be much slower vs. vectorized TruncatedBijector
(and UnivariateDistribution
should be implementing the minimum
and maximum
anyways, no?)
I can check again, but I don't think it works when the array of distributions is heterogeneous.
Ok, it actually seems to work... I can't reproduce the behavior I saw before. I'll just assume I did something wrong!
As noted in the code, the implementation of
bijector
for heterogeneous product distributions doesn't make sense. Why is not justStacked(map(bijector, d.v))
?Here is my fixed version: