避免在 Log10 操作中收到警告消息 [重复]

Question

如何避免在执行以下代码时收到警告消息？

from numpy import inf
a = np.array([0.2,0.2,0.3])
b = np.array([0.3,0.0,0.1])
a = np.log10(a)
b = np.log10(b)
b[b == -inf] = 0
a[a == -inf] = 0

calc = a+b

收到的警告信息：

divide by zero encountered in log10

代码应产生以下结果：

array([-1.22184875, -0.69897   , -1.52287875])

Answer 1

你可以用np.seterr去掉这个消息：

from numpy import inf
a = np.array([0.2,0.2,0.3])
b = np.array([0.3,0.0,0.1])
with np.seterr(divide = 'ignore'):
    a = np.log10(a)
    b = np.log10(b)
b[b == -inf] = 0
a[a == -inf] = 0
    
calc = a+b

如果您不想计算 log(0)，则需要像在欺骗目标中一样解决它：

calc = np.zeros_like(a)
a_ = a.astype(bool)   #False when 0
b_ = b.astype(bool)
np.add.at(calc, np.nonzero(a_), np.log10(a[a_]))
np.add.at(calc, np.nonzero(b_), np.log10(b[b_]))

这两个仅消除来自a, b == 0的错误，而不是来自a, b < 0的错误