The precision-recall curve plots precision against recall for all thresholds. It is more informative than the ROC curve when the positive class is rare, because it does not depend on true negatives. The area under the curve (average precision) summarizes it.
Example
precision_recall_curve(y_test, proba)