Pandas Series安全除法：替换除零inf为0时触发AttributeError

问题场景 #

实现了生成随机值Series的函数，每个元素是包含n个随机数的numpy数组。将两个该类型Series相除后，尝试用series1.divide(series2).replace(np.inf, 0)把除零产生的inf替换为0，除法操作正常，但调用replace时触发以下错误：

AttributeError: 'bool' object has no attribute 'to_numpy'

相关代码 #

生成随机值的函数：

def _draw_random_values(means: pd.Series,
                        standard_deviations: pd.Series, n: int = 10) -> pd.Series:
    return pd.Series([np.random.normal(mean, error, n)
                      for mean, error in zip(means, standard_deviations)])

Series结构示例 #

series1
0    [10.326329680446323, 10.341377563809141, 10.69...
1    [18.455738795462082, 20.24284540291898, 16.980...
2    [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, ...
dtype: object

series2[0][0] = 0
series2
0    [0.0, -1.4639471828693384, 18.085228130080917,...
1    [3.503465289188653, 7.2015882291641535, 13.146...
2    [7.520563427232638, 8.47603656244819, 14.34839...
dtype: object

除法结果（正常） #

series1.divide(series2)
0    [inf, -7.064037340158698, 0.5916429145326823, ...
1    [5.267852617925077, 2.810886259914426, 1.29171...
2    [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, ...

完整报错回溯 #

Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/util/_decorators.py", line 331, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/series.py", line 5380, in replace
    return super().replace(
           ^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/util/_decorators.py", line 331, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/generic.py", line 7280, in replace
    new_data = self._mgr.replace(
               ^^^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/internals/managers.py", line 467, in replace
    return self.apply(
           ^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/internals/managers.py", line 347, in apply
    applied = getattr(b, f)(**kwargs)
              ^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/internals/blocks.py", line 593, in replace
    mask = missing.mask_missing(values, to_replace)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/kmaguire/source/platf0rm-api/env/lib/python3.11/site-packages/pandas/core/missing.py", line 98, in mask_missing
    new_mask = new_mask.to_numpy(dtype=bool, na_value=False)
               ^^^^^^^^^^^^^^^^^
AttributeError: 'bool' object has no attribute 'to_numpy'

环境版本：pandas1.5.1，numpy1.23.4

原因分析 #

当前Series的dtype为object，每个元素是独立的numpy数组。Pandas 1.5.1的replace方法在处理这种嵌套数组的object类型Series时，内部逻辑存在缺陷：它会错误地将数组整体与np.inf比较，生成单个布尔值而非对应数组元素的掩码数组，后续代码试图对这个布尔值调用to_numpy方法，从而触发报错。

解决方案 #

方案1：用apply遍历每个数组替换inf #

对Series的每个元素（numpy数组）单独处理，使用np.where精准替换数组内的inf：

div_result = series1.divide(series2)
final_result = div_result.apply(lambda arr: np.where(np.isinf(arr), 0, arr))

这种方法直接针对每个数组操作，避开Pandasreplace方法的bug，适用于不想修改原有数据结构的场景。

方案2：重构数据结构为DataFrame #

修改生成函数，直接返回DataFrame而非元素为数组的Series，这样Pandas能正常处理数值型数据的替换：

def _draw_random_values(means: pd.Series,
                        standard_deviations: pd.Series, n: int = 10) -> pd.DataFrame:
    return pd.DataFrame([np.random.normal(mean, error, n)
                      for mean, error in zip(means, standard_deviations)])

# 使用方式
df1 = _draw_random_values(means, stds)
df2 = _draw_random_values(other_means, other_stds)
final_result = df1.divide(df2).replace(np.inf, 0)

DataFrame的每列都是数值类型，Pandas的replace方法能正确识别并替换所有inf值，是更规范的处理方式。

方案3：升级Pandas版本（可选） #

该bug在Pandas的后续版本（如1.5.3及以上、2.x系列）中已被修复，若允许升级依赖包，可直接升级后使用原有的replace代码：

pip install --upgrade pandas>=1.5.3

内容的提问来源于stack exchange，提问作者Kevin Kraft