Skip to content

Extend soft evidence to "all things considered"

Jan Pöppel requested to merge jpoeppel:factorTests into master

Created by: jpoeppel

The old (naive) soft evidence was equivalent to the "nothing else considered" case according to Darwiche in "Modeling and Reasoning with Bayesian Networks". This PR extends the factorTree set_evidence method to also allow the specification of soft evidence according to the "all things considered" method Darwiche mentions in the book.

Furthermore, this PR removes some limitations on factor division, by realising it as inversion and multiplication (will still need to profile the performance of this approach) and performs some further cleaning up.

Merge request reports