Scikit-learn平均精确率评分输入形状异常及PR曲线绘制求助

嘿，我之前也踩过这个坑！咱们来一步步搞清楚问题出在哪，怎么解决。

问题根源 #

你用clf.predict_proba(test_set)得到的y_score是一个二维数组，形状是(样本数量, 类别数量)——每个样本对应所有类别的预测概率。但average_precision_score在默认的二分类场景下，需要的是一维数组（只保留正类的预测概率）；如果是多分类/多标签任务，也需要调整输入格式才能匹配。

分场景解决方案 #

情况1：二分类任务（最常见）

这时候你只需要从predict_proba的结果中提取正类的概率列即可。假设LabelEncoder把你的正类编码成了1，代码修改如下：

import matplotlib.pyplot as plt
from sklearn import preprocessing
from sklearn.metrics import average_precision_score, precision_recall_curve

lbl_enc = preprocessing.LabelEncoder()
labels = lbl_enc.fit_transform(test_tags)

# 只保留正类的概率得分（[:, 1]表示取第二列，对应编码为1的类别）
y_score = clf.predict_proba(test_set)[:, 1]

average_precision = average_precision_score(labels, y_score)
print('Average precision-recall score: {0:0.2f}'.format(average_precision))

precision, recall, _ = precision_recall_curve(labels, y_score)
plt.step(recall, precision, color='b', alpha=0.2, where='post')
plt.fill_between(recall, precision, alpha=0.2, color='b')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.title(f'Precision-Recall Curve (AP={average_precision:.2f})')
plt.show()

情况2：多分类任务

如果你的任务是多分类，需要先把标签二值化，同时指定average参数来计算平均精确率：

import matplotlib.pyplot as plt
from sklearn import preprocessing
from sklearn.metrics import average_precision_score, precision_recall_curve

lbl_enc = preprocessing.LabelEncoder()
labels = lbl_enc.fit_transform(test_tags)
# 将标签二值化，每个类别对应一列
labels_binarized = preprocessing.label_binarize(labels, classes=lbl_enc.classes_)
n_classes = labels_binarized.shape[1]

y_score = clf.predict_proba(test_set)
# 选择合适的平均方式，比如'macro'（对每个类别单独计算后取平均）
average_precision = average_precision_score(labels_binarized, y_score, average='macro')
print(f'Average precision-recall score (macro): {average_precision:.2f}')

# 绘制每个类别的PR曲线
plt.figure()
for i in range(n_classes):
    precision, recall, _ = precision_recall_curve(labels_binarized[:, i], y_score[:, i])
    plt.step(recall, precision, alpha=0.2, where='post',
             label=f'Class {lbl_enc.classes_[i]} (AP={average_precision_score(labels_binarized[:, i], y_score[:, i]):.2f})')

plt.fill_between(recall, precision, alpha=0.2, color='b')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.ylim([0.0, 1.05])
plt.xlim([0.0, 1.0])
plt.title('Multi-class Precision-Recall Curve')
plt.legend(loc="lower left")
plt.show()

情况3：多标签任务

如果你的test_tags是多标签（每个样本可以属于多个类别），那你需要用MultiLabelBinarizer代替LabelEncoder，确保标签和y_score的形状匹配：

import matplotlib.pyplot as plt
from sklearn.preprocessing import MultiLabelBinarizer
from sklearn.metrics import average_precision_score, precision_recall_curve

mlb = MultiLabelBinarizer()
labels = mlb.fit_transform(test_tags)

y_score = clf.predict_proba(test_set)
average_precision = average_precision_score(labels, y_score, average='micro')
print(f'Average precision-recall score (micro): {average_precision:.2f}')

# 绘制PR曲线
precision, recall, _ = precision_recall_curve(labels.ravel(), y_score.ravel())
plt.step(recall, precision, color='b', alpha=0.2, where='post')
plt.fill_between(recall, precision, alpha=0.2, color='b')
plt.xlabel('Recall')
plt.ylabel('Precision')
plt.title(f'Multi-label Precision-Recall Curve (AP={average_precision:.2f})')
plt.show()

关键提醒 #

• 一定要确认你的任务类型（二分类/多分类/多标签），不同场景下输入格式要求不同
• precision_recall_curve的输入要求和average_precision_score一致，所以修改y_score的格式后，曲线绘制部分也能正常运行

内容的提问来源于stack exchange，提问作者tanmald