Closed dakota0064 closed 3 months ago
Things we know about the marriage bug
The income bug has been resolved, stemming from an issue in the masking during the embedding subset creation.
The marriage bug has also been resolved. We were sampling by individuals, not by pairs, so while we had an embedding overlap of 1,500 individuals per year we still had very few cases where both members of a pair had a valid embedding.
We're seeing bugs in both of these tasks when running one_model_per_year_variable_prediction. Income is showing as all 0s, and marriage is throwing an error because there isn't any overlap with the yearly embeddings / background.