How is LeaderBoard metric calculated?

In the problem description Home page, categorical crossentropy is put up as performance metrics and in submission page, evaluation metric is binary crossentropy. I guess since the competition is mutli-class classification categorical crossentropy is widely preferred . Kindly clarify or correct me if i’m wrong!!!

For multi-class classification, categorical cross_entropy and log_loss are same.
The evaluation metric is not for binary, as a fact it is for multi-class.
for binary the log_loss becomes:

-y*log(P(x)) - (1-y)*log(P(x))

more information:
sklearn-logloss-module

1 Like