This tags a stable(ish) version of Model 3 as described in the "GenDB - HIRM with PClean Entity Resolution" doc.
The schema is the same as the original HIRM, except that additional distributions and data types are supported (a bigram distribution over strings, a normal distribution over floats, categorical distributions over ints/strings, and a few others, in addition to the Bernoulli distribution in the original HIRM code). Most of these are specified in the schema simply as the name of the distribution (e.g. "normal"
) but a couple are more complicated -- see util_distribution_variant_test.cc
for examples.