probcomp / PClean

A domain-specific probabilistic programming language for scalable Bayesian data cleaning
216 stars 32 forks source link

Document the role of proposal dummy values #14

Open marcoct opened 3 years ago

marcoct commented 3 years ago

This text implies that ProposalDummyValues are only used for distributions that have infinite (and discrete) support:

https://github.com/probcomp/PClean/blob/master/src/distributions/distributions.jl#L10-L14

But StringPrior has finite support (there is a maximum length), and it implements discrete_proposal_dummy_value:

https://github.com/probcomp/PClean/blob/master/src/distributions/string_prior.jl#L23-L25

alex-lew commented 3 years ago

Good question -- it's really for support that's too large to tractably enumerate, not just infinite support. It is used whenever preferred values are used (it's the "other" token from the PClean paper).