Clean attributes should contribute the probability in the tensor.

[zaqthss]

All the attributes in the database should be used to compute the probability and insert into the feature.
In cooccurrencefeaturizer.py, get_query function, line 91, only attribute in dirty_cells_attributes are recorded as a feature. In insert_to_tensor function, line 62 and 64, the clean attributes then will not contribute any probability in the tensor since the feature is -1.