我们可以用sklearn.metrics中的precision_recall_curve()和auc()计算出PR-AUC，然后用matplotlib.pyplot画出PR-curve：
代码语言：javascript复制from sklearn.metrics import precision_recall_curve, auc
clf = LogisticRegression()
clf.fit(X_train, y_train)
y_pred_proba = clf.predict_proba(X_test)[::,1]
clf.fit(X_train_e, y_train_e)
y_pred_proba_e = clf.predict_proba(X_test_e)[::,1]
precision, recall, thresholds = precision_recall_curve(y_test, y_pred_proba)
precision_e, recall_e, thresholds_e = precision_recall_curve(y_test_e, y_pred_proba_e)
pr_auc = auc(recall, precision)
pr_auc_e = auc(recall_e, precision_e)
plt.plot(recall,precision, color = 'blue')
plt.plot(recall_e,precision_e, color = 'red')
plt.title('Precision/Recall Curve')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.legend(loc="upper right", labels = ['Full model {}'.format(round(pr_auc, 2)), 'Expression only model {}'.format(round(pr_auc_e, 2))])
plt.show()
值得注意的是，对于特别不平衡的样本，虽然ROC-AUC可能会很好看，但是PR-AUC多半很一般，甚至很不好，上采样和下采样是非常有必要的，另外不要被ROC_AUC所蒙蔽。
欢迎关注！
