将值从一个 dataframe 切片复制到另一个：使用“IndexSlice”的多索引 pandas 数据帧的切片是否总是一致地排序？

Question

Context

Say I have a multi-indexed dataframe as follows:

import numpy as np
import pandas as pd

arrays = [
    ["bar", "bar", "baz", "baz", "foo", "foo", "qux", "qux"],
    ["one", "two", "one", "two", "one", "two", "one", "two"],
]
tuples = list(zip(*arrays))
index = pd.MultiIndex.from_tuples(tuples, names=["first", "second"])
data = np.array([
    [1, 2],
    [3, 4],
    [5, 6],
    [7, 8],
    [9, 10],
    [11, 12],
    [13, 14],
    [15, 16],
])
df = pd.DataFrame(data, index=index, columns=('a', 'b'))

which looks something like this:

               a   b
first second        
bar   one      1   2
      two      3   4
baz   one      5   6
      two      7   8
foo   one      9  10
      two     11  12
qux   one     13  14
      two     15  16

I would like to copy the values of column a for the first index level bar into the same column for the first index level qux , aligned on the second level of the index (here called second ). In other words, I would like to obtain the following dataframe from the one above:

               a   b
first second        
bar   one      1   2
      two      3   4
baz   one      5   6
      two      7   8
foo   one      9  10
      two     11  12
qux   one      1  14  # <-- column a changed to match first = bar for second = one
      two      3  16  # <-- column a changed to match first = bar for second = two

I understand based on the answer given to this question I can accomplish this by using pd.IndexSlice in conjunction with .loc and .values as follows:

df.loc[pd.IndexSlice['qux', :], 'a'] = df.loc[pd.IndexSlice['bar', :], 'a'].values

I don't intuitively like this (perhaps/probably unjustifiably) as it's not immediately clear to me if the values will always be aligned on the second index level or not:

Question:

Can I guarantee that the above assignment (accessing using .values ) will always be aligned on the second level of the multi-index?

If not, is there a way of accomplishing what I'm trying to achieve?

Answer 1

No, it will not be aligned, because by using .value (which, by the way, is deprecated in favor of .to_numpy() ), which returns the underlying numpy array, you remove all index/column information, so alignment is not possible.

Here's one solution to preserve the alignment:

df.loc['qux', 'a'] = df.loc['qux', 'a'].index.map(df.loc['bar', 'a'].to_dict())```

Output:

>>> df
                 a   b
first second          
bar   two      1.0   2
      one      3.0   4
baz   one      5.0   6
      two      7.0   8
foo   one      9.0  10
      two     11.0  12
qux   one      3.0  14
      two      1.0  16