Feras Saad

bug in INDEPENDENT initialization, revise.

Something tells me that these capabilities should not be built-in to the inference engine, but rather an intermediate layer, between the inference backend and the user, which achieves the dependence...

With the `state.diagnostics` dictionary, should be able to fix this easily now.

@riastradh-probcomp yes, your interpretation is the one that lovecat uses. The reason that I put this issue in cgpm, and not in bayeslite, is how can we define the above...

> On the other hand, we could consistently interpret cgpm.logpdf(A_r = 42) to mean log Pr[A_\* = 42 | D, B_\* = b_r, X_\* = x_r] -- i.e., any time...

This issue is challenging because the Dim is never initialized with a partition but the parameters are one per cluster.

Also useful for computational considerations: http://www.stats.ox.ac.uk/~doucet/samsi_course.html

But the issue is that it also changes the `__ini__` interface of State and Engine, which is a big effort to change clients especially tests.

Temporary workaround is to allow unincorporate of the last rowid only, which means we do not have to worry about re-indexing the rowids as they are guaranteed to remain contiguous....

@axch implemented some forms of Geweke testing in bayeslite some time ago, perhaps he can weigh in on their usefulness to help decide whether it is worth keeping this ticket...