Add a test for a more complex schema + bug fixes. #229
Conversation
| // Samples and incorporates a value into all relations belonging to class_name | ||
| // (including class attributes and noisy observations of ancestor class | ||
| // attributes). Recursively calls itself on base relations. | ||
| void GenDB::sample_and_incorporate_for_class(std::mt19937* prng, |
There was a problem hiding this comment.
Just curious -- will this work when class_name is the name of the observation class? Because if so, we could use it to slightly simplify the implementation of make_pclean_sample in pclean_lib.
There was a problem hiding this comment.
Good point, this should work for the observation class too.
cxx/gendb.cc
Outdated
| // equals class_item. | ||
| // equals class_item. For singleton references, recursively unincorporates | ||
| // from the reference class as well. | ||
| // TODO there is a known issue where an entity with a value that is perfectly |
There was a problem hiding this comment.
Can you create a GitHub issue for this with more details? In particular, I'm not sure what perfect correlation means here.
There was a problem hiding this comment.
I rethought this and I don't think it's actually an issue (if the entities don't share a DAG path, the "correlated" one won't change after the Gibbs step, even if it technically should have been unincorporated in intermediate states -- by "correlated" I just mean that Reference A takes value X iff Reference B takes value Y. I deleted the comment.
cxx/gendb.cc
Outdated
| @@ -304,7 +356,7 @@ double GenDB::unincorporate_reference_relation( | |||
| std::unordered_map<T_items, ObservationVariant, H_items>>& | |||
| stored_value_map) { | |||
| // We can stop unincorporating from base relations if the index of | |||
| // the domain of interest is greater than the number of domains. | |||
| // the domain of interest is greater than the max index. | |||
There was a problem hiding this comment.
max index of what?
There was a problem hiding this comment.
I meant the max index of the vector of items, so the length minus one. I rephrased the comment.
emilyfertig
force-pushed
the
100324-emilyaf-gendb-complex-schema
branch
from
567151d
to
17af275
Compare
Also renames domain_crps -> entity_crps to disambiguate from domain.crp in the IRMs.