Handle PARSynthesizer model if sequence_index is missing #114
Conversation
lajohn4747
requested review from
frances-h and
amontanez24
and removed request for
a team
| @@ -181,7 +182,8 @@ def assemble_sequences( | |||
| groupby_columns = entity_columns[0] if len(entity_columns) == 1 else entity_columns | |||
| for _, sequence in data.groupby(groupby_columns): | |||
| sequence.drop(entity_columns, axis=1, inplace=True) | |||
| if context_columns: | |||
| missing_columns = [col for col in context_columns if col not in sequence.columns] | |||
| if context_columns and not missing_columns: | |||
There was a problem hiding this comment.
should we check the other columns that are in sequence instead of skipping over? Or is the fake column the only one in context_columns?
There was a problem hiding this comment.
Why are missing context columns being passed to PAR in the first place? My understanding is that the UUID column gets added so that we have a dummy column for the context synthesizer. Presumably it should have either (1) added a dummy constant column to the data or (2) not be passed along to PAR at all.
Looking into this more, I think the problem is actually that we're adding the UUID column to |
Why does the UUID column need to be added for modeling purposes? Seems like the issue is resolved and all tests pass (with the exception of a unit test checking for the added column) when I remove the the added UUID column, so I am not sure if it is still needed. |
I think the problem is that we can't fit on an empty dataframe, so when there's no context columns we have to add a dummy column to correctly create the context model without erroring out. |
|
Moved fix to https://github.com/sdv-dev/SDV/pull/2019/files |
resolves sdv-dev/SDV#1972
CU-86b08wr44
When sequence index is missing, par.py adds a constant column to allow for modeling as seen here. The added context column does not exist in the data though causing
KeyErrors. Added a check to prevent failures.