When probcomp/crosscat#64 is merged (and logpdf is adjusted to call the correct thing), we should have all the machinery needed to run Geweke tests against our crosscat implementation, as least for one-row data sets. We should do that.
I expect this test to be sensitive enough to distinguish between a correct implementation of logpdf and passing a bulk query to simple_predictive_sample and multiplying the results together (summing in log domain).
When probcomp/crosscat#64 is merged (and logpdf is adjusted to call the correct thing), we should have all the machinery needed to run Geweke tests against our crosscat implementation, as least for one-row data sets. We should do that.