Python numpy.random.choice : ValueError: Fewer non-zero entries in p than size

Question

我想根据给定行的 prob 指定的概率分布随机选择样本点。但是，当我调用np.random.choice时，我收到错误ValueError:p中非零条目少于大小。 size 到底是什么意思？我还看了implementation但我不明白。谢谢你的帮助!!

import numpy as np

# prob is a numpy array of shape (14, 6890)
all_zero = np.where(prob.max(1) < 1e-6)[0] # find indices of rows where all values are smaller
prob[all_zero] = 1 / prob.shape[1] # fill those rows uniformly
prob /= prob.sum(axis=1, keepdims=True)
# ... somewhere later inside a method
for j in range(14):
    sample = np.random.choice(6890, 4, replace=False, p=prob[j]) # error occurs here

Answer 1

问题出在您使用 np.random.choice 时，您要求选择 4 个条目而不重用值 (replace=False)在包含 6890 个条目的数组中非空值少于 4 个，例如:

>>> np.random.choice(5, 1, replace=False, p=[0, 0, 0, 0.6, 0.4])
array([4])

>>> np.random.choice(5, 4, replace=False, p=[0, 0, 0, 0.6, 0.4])
Traceback (most recent call last):
  File "<input>", line 1, in <module>
    np.random.choice(5, 4, replace=False, p=[0, 0, 0, 0.6, 0.4])
  File "mtrand.pyx", line 826, in numpy.random.mtrand.RandomState.choice
ValueError: Fewer non-zero entries in p than size

>>> np.random.choice(5, 4, replace=True, p=[0, 0, 0, 0.6, 0.4])
array([3, 3, 4, 3])

因此解决方案取决于您的需要，您要么确保有更多的非空值，要么在随机选择中启用替换。

作为引用，numpy.random.choice 的文档:

• https://docs.scipy.org/doc/numpy-1.15.0/reference/generated/numpy.random.choice.html