问题

我在Pandas中有一列Numpy数组，类似于：
```python
   col1 col2    col3
0     1    a    None
1     2    b  [2, 4]
2     3    c    None

[2, 4]实际上是np.array([2, 4])。现在我需要填充缺失值，我有一个用于填充的数组列表。例如：

vals_to_impute = [np.array([1, 2]), np.array([1, 4])]

我尝试了以下操作：

mask = col3.isna()
df.loc[mask, "col3"] = vals_to_impute

但是出现了错误：

ValueError: Must have equal len keys and value when setting with an ndarray

我尝试将其转换为Numpy数组，提取列等，但都没有成功。是否实际上可以使用矢量化操作来设置它，还是必须手动循环？

<details>
<summary>英文:</summary>

I have a column of Numpy arrays in Pandas, something like:

col1 col2 col3
0 1 a None
1 2 b [2, 4]
2 3 c None

The `[2, 4]` is really `np.array([2, 4])`. Now I need to impute the missing values, and I have a list of arrays for that. For example:

vals_to_impute = [np.array([1, 2]), np.array([1, 4])]

I tried:

mask = col3.isna()
df.loc[mask, "col3"] = vals_to_impute

This results in error:

ValueError: Must have equal len keys and value when setting with an ndarray

I tried converting to Numpy array, extracting column etc., nothing worked. Is it actually possible to set this in a vectorized operation, or do I have to do a manual loop?

</details>


# 答案1
**得分**: 3

我使用`pd.Series`而不是列表来完成了这个任务。我还必须为这个Series输入一个索引，以确保插入正确。也许有更简单的方法。

```python
df = pd.DataFrame({
    "col1": [1, 2, 3],
    "col2": ["a", "b", "c"],
    "col3": [None, np.array([2, 4]), None]
})

mask = df["col3"].isna()
vals_to_impute = pd.Series(
    [np.array([1, 2]), np.array([1, 4])],
    index=mask[mask].index
)

df.loc[mask, "col3"] = vals_to_impute

print(df)

输出：

   col1 col2    col3
0     1    a  [1, 2]
1     2    b  [2, 4]
2     3    c  [1, 4]

df = pd.DataFrame({
    "col1": [1, 2, 3],
    "col2": ["a", "b", "c"],
    "col3": [None, np.array([2, 4]), None]
})

mask = df["col3"].isna()
vals_to_impute = pd.Series(
    [np.array([1, 2]), np.array([1, 4])],
    index=mask[mask].index
)

df.loc[mask, "col3"] = vals_to_impute

print(df)

   col1 col2    col3
0     1    a  [1, 2]
1     2    b  [2, 4]
2     3    c  [1, 4]

答案2

得分: 2

一种使用短循环的选项：

mask = df['col3'].isna()
vals = iter(vals_to_impute)

for idx in df.index[mask]:
    df.at[idx, 'col3'] = next(vals, None)

这也可以通过类似于此处提供的方法解决（https://stackoverflow.com/a/76641469/16343464），利用底层 numpy 数组的共享内存：

mask = df['col3'].isna()
arr = df['col3'].to_numpy()
arr[np.where(mask)] = vals_to_impute

修改后的 DataFrame：

   col1 col2    col3
0     1    a  [1, 2]
1     2    b  [2, 4]
2     3    c  [1, 4]

使用的输入：

df = pd.DataFrame({'col1': [1, 2, 3],
                   'col2': ['a', 'b', 'c'],
                   'col3': [None, np.array([2, 4]), None]})

vals_to_impute = [np.array([1, 2]), np.array([1, 4])]

mask = df['col3'].isna()
vals = iter(vals_to_impute)

for idx in df.index[mask]:
    df.at[idx, 'col3'] = next(vals, None)

mask = df['col3'].isna()
arr = df['col3'].to_numpy()
arr[np.where(mask)] = vals_to_impute

   col1 col2    col3
0     1    a  [1, 2]
1     2    b  [2, 4]
2     3    c  [1, 4]

df = pd.DataFrame({'col1': [1, 2, 3],
                   'col2': ['a', 'b', 'c'],
                   'col3': [None, np.array([2, 4]), None]})

vals_to_impute = [np.array([1, 2]), np.array([1, 4])]

答案3

得分: 1

不确定你的问题实际在哪里，因为你只展示了你代码的一部分。。

对我来说，这正如预期的那样工作：

import pandas as pd
import numpy as np

data = [None, None, np.array([2, 2]), None, None]
df = pd.DataFrame(dict(data=data))

mask = df["data"].isna()
maskdata = [np.array([i, i]) for i in mask[mask].index]

df.loc[mask, "data"] = maskdata

# print(df)
# >>>     data
# >>> 0  [0, 0]
# >>> 1  [1, 1]
# >>> 2  [2, 2]
# >>> 3  [3, 3]
# >>> 4  [4, 4]

请注意，我已经按照你的要求，只返回翻译好的代码部分。

import pandas as pd
import numpy as np

data = [None, None, np.array([2, 2]), None, None]
df = pd.DataFrame(dict(data = data))

mask = df["data"].isna()
maskdata = [np.array([i, i]) for i in mask[mask].index]

df.loc[mask, "data"] = maskdata

# print(df)
# >>>     data
# >>> 0  [0, 0]
# >>> 1  [1, 1]
# >>> 2  [2, 2]
# >>> 2  [3, 3]
# >>> 2  [4, 4]

答案4

得分: 1

根据我的理解，问题出现在loc对基于索引匹配的值产生影响。

以你的示例为例：

df = pd.DataFrame({"col1":[1,2,3], 
                   "col2":["a","b","c"], 
                   "col3":[None, np.array([2,4]), None]})

如果你看一下：

mask = df.col3.isna()
mask

实际上你会得到：

0     True
1    False
2     True
Name: col3, dtype: bool

因此，尽管你只选择了具有“True”的行，但索引仍然具有长度为3。此外，你的列表vals_to_impute没有索引，因此Pandas不知道如何影响值。

一个快速的修复方法可能是：

mask = df[df.col3.isna()].index
vals_to_impute = pd.Series([np.array([1, 2]), np.array([1, 4])], index=mask)
df.loc[mask,"col3"] = vals_to_impute

注意：

• 可能有一种更合适的Pandas方法来做这个。
• 在DataFrame列中使用nd.arrays在我所知道的情况下相当不常见，请注意你可能会在以后遇到其他问题。

df = pd.DataFrame({"col1":[1,2,3], 
                   "col2":["a","b","c"], 
                   "col3":[None, np.array([2,4]), None]})

mask = df.col3.isna()
mask

0     True
1    False
2     True
Name: col3, dtype: bool

mask = df[df.col3.isna()].index
vals_to_impute = pd.Series([np.array([1, 2]), np.array([1, 4])], index=mask)
df.loc[mask,"col3"] = vals_to_impute