Is there someone else that experienced this with the new data?
I'm getting significant less cases of person_ids in the true positive class in contrast to the fast lane data from the previous round and this affects my model's overall AUC metric. Was there something wrong with the previous dataset or is it with this one? It could be an error from my side though...
Created by peteoc Hi @peteoc,
The synthetic data is not meant to be used to estimate model performance, it should only give you an idea of how the UW data is formatted. If you are finding that the previous round's fast lane data is more useful for model performance estimation, you are welcome to use it to design your model on your local environment, but be sure to test it on the current fast lane data before you submit.
Thanks,
Tim