I tried different approaches in order to reduce this gap but I think the test set is different from the train set
By the way, when I summed the logloss error for the predicted columns independently* my CV score was similar to the LB score. I am not sure if it's a coincidence or may be Zindi is miscalculating the error.
*sum(logloss(predicted_probability_class_i, 1 if class_i else0 ))
Yes, I have the same problem getting 0.75 CV
I tried different approaches in order to reduce this gap but I think the test set is different from the train set
By the way, when I summed the logloss error for the predicted columns independently* my CV score was similar to the LB score. I am not sure if it's a coincidence or may be Zindi is miscalculating the error.
*sum(logloss(predicted_probability_class_i, 1 if class_i else0 ))
I tested the same idea and i get the same score like the lb
How to check the CV
Hi everyone. Thank you for flagging this potential issue. We are looking into it now. We will keep you updated.
No traction on this yet, I hope this issue is sorted out early enough.
Morning everyone, thank you for your patience. We have identified the issue and it should be resolved in a week.
is this solved ? looks like no
yes
indeed it is now.