Pandas多索引Series添加行并归入对应外层索引组（禁用sort_index）

需求：给多索引Series新增一行，需将该行归入指定的外层索引组，且必须保留原有索引的字母顺序，因此不能使用df.sort_index()。

原代码 #

import pandas as pd
import numpy as np

categories = {"A":["c", "b", "a"] , "B": ["a", "b", "c"], "C": ["a", "b", "d"] }
array = []
expected_fields = []
for key, value in categories.items():
    array.extend([key]* len(value))
    expected_fields.extend(value)
    
arrays = [array ,expected_fields]

tuples = list(zip(*arrays))
index = pd.MultiIndex.from_tuples(tuples)
df =  pd.Series(np.random.randn(9), index=index)
df["A", "d"] = 2

print(df)

当前输出 #

A  c    0.887137
   b   -0.105262
   a   -0.180093
B  a   -0.687134
   b   -1.120895
   c    2.398962
C  a   -2.226126
   b   -0.203238
   d    0.036068
A  d    2.000000 <------------
dtype: float64

期望输出 #

A  c    0.887137
   b   -0.105262
   a   -0.180093
   d    2.000000  <--------------
B  a   -0.687134
   b   -1.120895
   c    2.398962
C  a   -2.226126
   b   -0.203238
   d    0.036068
dtype: float64

解决方案 #

直接通过索引赋值会把新行追加到Series末尾，要让新行归入对应外层索引组，可通过以下两种方式实现：

方法1：拆分原Series后合并新行 #

import pandas as pd
import numpy as np

categories = {"A":["c", "b", "a"] , "B": ["a", "b", "c"], "C": ["a", "b", "d"] }
array = []
expected_fields = []
for key, value in categories.items():
    array.extend([key]* len(value))
    expected_fields.extend(value)
    
arrays = [array ,expected_fields]
tuples = list(zip(*arrays))
index = pd.MultiIndex.from_tuples(tuples)
df =  pd.Series(np.random.randn(9), index=index)

# 创建要添加的新行
new_row = pd.Series([2], index=pd.MultiIndex.from_tuples([("A", "d")]))

# 拆分原Series为A组和其他组
a_group = df.loc["A"]
other_groups = df.drop("A")

# 合并A组与新行，再合并其他组
updated_a = pd.concat([a_group, new_row])
result = pd.concat([updated_a, other_groups])

print(result)

方法2：通过重新索引插入指定位置 #

如果需要精确控制新行在组内的位置（比如不是组末），可以先调整索引列表再重新索引：

import pandas as pd
import numpy as np

categories = {"A":["c", "b", "a"] , "B": ["a", "b", "c"], "C": ["a", "b", "d"] }
array = []
expected_fields = []
for key, value in categories.items():
    array.extend([key]* len(value))
    expected_fields.extend(value)
    
arrays = [array ,expected_fields]
tuples = list(zip(*arrays))
index = pd.MultiIndex.from_tuples(tuples)
df =  pd.Series(np.random.randn(9), index=index)

# 获取原索引的列表形式
index_list = list(df.index)
# 找到A组最后一个元素的位置
last_a_pos = [i for i, idx in enumerate(index_list) if idx[0] == "A"][-1]
# 在A组末尾插入新索引
index_list.insert(last_a_pos + 1, ("A", "d"))

# 重新索引并赋值新行
df = df.reindex(index_list)
df.loc[("A", "d")] = 2

print(df)

以上两种方法都能保留原有索引的顺序，同时将新行归入指定的外层索引组。

内容的提问来源于stack exchange，提问作者saul santos