Could some explain how Zindi computes the scores of the private LB ? We are only given the public test data, and Zindi don't have the models we build, so how can they compute the score on the private LB? Just noted a changed on the scores and the LB? Or are they still having a problem with the log_loss ?
The test set you make predictions on contains both the private and public LB data. Upon submission, only the public part of the test is used to get your score. The private LB predictions is done at the close of competition
I see, thanks! I didn't know the test set contained both public and private sets.