在sklearn predict_proba()方法中，通常我们只查看概率最高的方法。如何输出前n个类（n> 1）的概率？

例如，输出predict_proba()是这样的，我如何返回最高的2个概率及其关联类？

result_prob = clf.predict_proba(X_test)

返回：

array([

2.55420153e-02, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

3.41739673e-02, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 2.11688875e-05, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 8.02579585e-01, 0.00000000e+00,

0.00000000e+00, 1.37978949e-02, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 1.15640553e-02, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 6.76391638e-02,

9.06030431e-03, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 3.56218448e-02, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 0.00000000e+00])

在这种情况下，应该返回概率为8.02579585e-01和的类6.76391638e-02。

解决办法：这实际上是一个Numpy问题; 你可以使用np.argpartition：

import numpy as np

x =np.array([

2.55420153e-02, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

3.41739673e-02, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 2.11688875e-05, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 8.02579585e-01, 0.00000000e+00,

0.00000000e+00, 1.37978949e-02, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 1.15640553e-02, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 6.76391638e-02,

9.06030431e-03, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 3.56218448e-02, 0.00000000e+00,

0.00000000e+00, 0.00000000e+00, 0.00000000e+00])

k = 2 # top-k

ind = np.argpartition(x, -k)[-k:]

x[ind]

结果：

array([0.06763916, 0.80257959])

根据要求，各个班级在ind：

ind

# array([27, 14])