When using the following TransformedDistribution my NUTS sampler is initially sampling values outside of the support expected for a Beta distribution. I believe this is because transformed_dist.support is Real() when it should realistically be numpyro.distributions.constraints._Interval(1.0, 2.0).
1) Is this expected behavior?
2) would this cause cause issues with initial samples from NUTS being out of expected range?
Running the following code displays the issue I am running into.
When using the following
TransformedDistribution
my NUTS sampler is initially sampling values outside of the support expected for a Beta distribution. I believe this is becausetransformed_dist.support
isReal()
when it should realistically benumpyro.distributions.constraints._Interval(1.0, 2.0)
.1) Is this expected behavior? 2) would this cause cause issues with initial samples from NUTS being out of expected range?
Love the package, thanks guys!