Train.csv has labels, Test.csv does not

ya... i like the challenge for sure, but getting a bit frustrating! what model type are you using? i'm using catboost exclusively as it handles the missing values in the Test.csv.

Really frustrating man!😅 Tbh I think I am not doing things right because if I were doing things right, my CV and LB should correlate.

I tried Lightgbm with folds but I trained monthly and aggregated the result (0.98AUC CV 0.88 AUC LB) - This was my best result from tree models.

I also tried a deep learning approached. Used a time series approach, masked some values in order to handle the missing values.

This gave my score right now: (AUC 0.914110206 LB, 0.99CV)

I am not focusing on F1 right now until I can correlate AUC and my CV.

on the Train.csv or the Test.csv? If that is on the Test.csv, that is really good!

I am stuck at 0.8 too.

For both actually. In my experiments, I aligned train and test to push to 0.88 AUC LB.

I decided to make each month a training data point. So instead of 1821, I had about 21852. Same was applied to test and I aggregated the results to get that submission.