AUC measures how well a model ranks cases: the chance that a real positive gets a higher score than a real negative. To do that, it needs the model's scores, not its final yes-or-no answers.
Task: write two_aucs(y_true, chance) and return [auc_from_chances, auc_from_labels], each rounded to 3 decimal places.
y_true holds 1 for positive and 0 for negative. Both appear.auc_from_chances is the ROC AUC computed from chance directly.auc_from_labels is the ROC AUC computed from hard labels: 1 where the chance is at or above 0.5, and 0 otherwise.The second number is what you get if you hand AUC the output of predict by mistake. Compare it with the first.