You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently, we have one-side truncated normal and cauchy implemented. But it would be nice to add a general class, which uses cdf method, for other distributions. There is a nice discussion here for context on relevant numerical issues.
Note we also have the one-sided general FoldedDistribution in Pyro, which is a cheap rsample-friendly qualitatively similar alternative to TruncatedDistribution. This doesn't get us any closer to literally translating Stan models, but maybe we can take license to approximately translate models and e.g. check that we're only slightly truncating, or not truncating past the median (since Folded and Truncated differ for severely truncated distributions).
Currently, we have one-side truncated normal and cauchy implemented. But it would be nice to add a general class, which uses
cdf
method, for other distributions. There is a nice discussion here for context on relevant numerical issues.cc @fritzo
The text was updated successfully, but these errors were encountered: