sklearn之average_precision_score计算返回NaN

问题描述

使用sklearn计算AP时，当label全是负标签时会返回NaN，例如：

>>> import numpy as np
>>> from sklearn.metrics import average_precision_score
>>> average_precision_score(np.array([0, 0, 0, 0, 0]), np.array([0.1, 0.1, 0.1, 0.1, 0.1]))

xxx/lib/python3.7/site-packages/sklearn/metrics/_ranking.py:864: RuntimeWarning: invalid value encountered in true_divide
  recall = tps / tps[-1]
nan

问题解决

参考average_precision_score does not return correct AP when all negative ground truth labels，这个Bug在scikit-learn==1.1.0之后的版本被修复，因此升级sklearn的版本即可。
具体方法：在Python (>= 3.8)下执行命令，安装新的版本

pip install scikit-learn==1.1.0

效果如下：

>>> import numpy as np
>>> from sklearn.metrics import average_precision_score
>>> average_precision_score(np.array([0, 0, 0, 0, 0]), np.array([0.1, 0.1, 0.1, 0.1, 0.1]))

xxx/lib/python3.8/site-packages/sklearn/metrics/_ranking.py:874: UserWarning: No positive class found in y_true, recall is set to one for all thresholds.
  warnings.warn(
-0.0

参考

1、average_precision_score does not return correct AP when all negative ground truth labels
2、FIX Fix recall in multilabel classification when true labels are all negative
3、scikit-learn 1.1.0