Since #905 was merged, we are storing the score of each individual choice/sample as part of the trace. (See here.) It looks like we can re-use these (rather than re-scoring) when computing MH transition probabilities, saving some work. i.e. here I guess we can replace choice.dist.score(choice.val) with choice.choiceScore.
Since #905 was merged, we are storing the score of each individual choice/sample as part of the trace. (See here.) It looks like we can re-use these (rather than re-scoring) when computing MH transition probabilities, saving some work. i.e. here I guess we can replace
choice.dist.score(choice.val)withchoice.choiceScore.