负类的 precision_recall_curve 导致精确度和召回率呈正相关

Question

我有一个逻辑回归模型来预测二进制输出（0 或 1）。我想了解0级的P/R，并生成相应的曲线。我使用这个代码：

clf = linear_model.LogisticRegression().fit(ohe_X_train, y_train)
# Predict labels on the one hot encoded test set
clf_predictions = clf.predict(ohe_X_test)
y_scores = clf.predict_proba(ohe_X_test)

class_of_interest = 0

# P/R based on precision_recall_fscore_support
precision, recall, fscore, support = precision_recall_fscore_support(y_test, clf_predictions, labels=[class_of_interest])

# P/R curve using precision_recall_curve
y_scores = clf.predict_proba(ohe_X_test)[:, class_of_interest]
precision_curve, recall_curve, thresholds = precision_recall_curve(y_test, y_scores)

plt.plot(recall_curve, precision_curve, marker='.')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.title('Precision-Recall Curve for Class 0')
plt.grid(True)
plt.show()

在这种情况下，PR 曲线随着召回率的增加而提高了精确度。我做错了什么？当 class_of_interest = 1 时它完美工作。

Answer 1

而不是

precision_curve, recall_curve, thresholds = precision_recall_curve(y_test, y_scores)

应该是

precision_curve, recall_curve, thresholds = precision_recall_curve(1-y_test, y_scores)