Fix roc_auc by inputing binary ground truth and continuous confident prediction
Fix other score by inputing binary ground truth and binary prediction
Major and minor reporting scores change from accuracy to recall. I suppose we want recall instead of accuracy.
When comparing 95 recall of non-hybrids, we may have the same predictions of different images, so instead of looping each image, we may want to loop the set of prediction (distinct values of prediction) to include the image instances with the same prediction during calculating the recall.
Add prc_auc
Add major and minor prc_auc and roc_auc
Leaderboard reporting score changes from roc_auc to prc_auc
run.shto run the code in local