Change to unbiased importance sampling estimator, easier to defend than the biased sampler I had been using
Remove the dependence of the log partition function derivative on the expectation method. This is so that in #225 I can restrict expectation to accept only tensor valued functions instead of allowing structure valued functions.
Two updates:
expectation
method. This is so that in #225 I can restrictexpectation
to accept only tensor valued functions instead of allowing structure valued functions.