sklearn 多类 roc auc 分数

sklearn multiclass roc auc score

如何在 sklearn 中获取 multi-class classification 的 roc auc 分数?

二进制

# this works
roc_auc_score([0,1,1], [1,1,1])

多class

# this fails
from sklearn.metrics import roc_auc_score

ytest  = [0,1,2,3,2,2,1,0,1]
ypreds = [1,2,1,3,2,2,0,1,1]

roc_auc_score(ytest, ypreds,average='macro',multi_class='ovo')

# AxisError: axis 1 is out of bounds for array of dimension 1

看了官方documentation,还是没能解决

roc_auc_score 在多标签情况下需要具有形状 (n_samples, n_classes) 的二进制标签指示符,这是回到 one-vs-all 时尚的方式。

要轻松做到这一点,您可以使用 label_binarize (https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.label_binarize.html#sklearn.preprocessing.label_binarize)。

对于您的代码,它将是:

from sklearn.metrics import roc_auc_score
from sklearn.preprocessing import label_binarize

# You need the labels to binarize
labels = [0, 1, 2, 3]

ytest  = [0,1,2,3,2,2,1,0,1]

# Binarize ytest with shape (n_samples, n_classes)
ytest = label_binarize(ytest, classes=labels)

ypreds = [1,2,1,3,2,2,0,1,1]

# Binarize ypreds with shape (n_samples, n_classes)
ypreds = label_binarize(ypreds, classes=labels)


roc_auc_score(ytest, ypreds,average='macro',multi_class='ovo')

通常,这里的 ypreds 和 yest 变成:

ytest
array([[1, 0, 0, 0],
       [0, 1, 0, 0],
       [0, 0, 1, 0],
       [0, 0, 0, 1],
       [0, 0, 1, 0],
       [0, 0, 1, 0],
       [0, 1, 0, 0],
       [1, 0, 0, 0],
       [0, 1, 0, 0]])

ypreds
array([[0, 1, 0, 0],
       [0, 0, 1, 0],
       [0, 1, 0, 0],
       [0, 0, 0, 1],
       [0, 0, 1, 0],
       [0, 0, 1, 0],
       [1, 0, 0, 0],
       [0, 1, 0, 0],
       [0, 1, 0, 0]])