This relies on #22. Once #22 is fixed, one can simply add a function for logprob based on rejection sampling to the ConstrainedGPM (similar to what is done a logpdf here).
Alternatively, this could be deterministically computed once #22 is fixed. That's more complicated and would require re-thinking conditioning in ClojureCat in general -- likely ending in making ClojureCat inference more similar to SPPL's inference.
This relies on #22. Once #22 is fixed, one can simply add a function for logprob based on rejection sampling to the ConstrainedGPM (similar to what is done a
logpdfhere).Alternatively, this could be deterministically computed once #22 is fixed. That's more complicated and would require re-thinking conditioning in ClojureCat in general -- likely ending in making ClojureCat inference more similar to SPPL's inference.