i think there is a problem a problem on how the log_loss is computed , simply because jus shuffling columns on submission file results into different log_loss score , nomarlly this should result give the same logloss score
can i get help on this......................................................?
It's not normal. We have make do with!
meaning.......?, column shuffling should have effect on log_loss?
No, the column shuffling shouldn't have effect on log_loss. The topic already came up in a discussion. Zindi knows. We have to do with!
And what was the solution , coz this issue results into misalignment between CV score and LB score........
I think the scoring is done using indexes and not column names, I get a scoring in the same order as the one I computed for a validation set if I put the columns in the order given in the submissions file.
I think better stick with the order they gave in the submissions file. :)