多级索引DataFrame处理：删NaN行，留0级索引并单独重置各1级索引

Got it, let's break down how to solve this problem exactly as you described—first cleaning up NaN rows while keeping your Level 0 index intact, then resetting the Level 1 index to start from 0 for each Level 0 group.

Step 1: Remove Rows with NaN Values #

First, we'll use dropna() to eliminate any rows containing NaN values. This preserves your original multi-index structure (including both Level 0 and Level 1) while only removing the problematic rows:

import pandas as pd

# Assume your DataFrame is named `df`
df_clean = df.dropna()

Step 2: Reset Level 1 Index for Each Level 0 Group #

The key here is to use groupby() on the Level 0 index, then generate a new sequential index starting at 0 for each group. The most efficient way to do this is with cumcount(), which creates a counter for each group:

# Remove the original Level 1 index temporarily
df_temp = df_clean.reset_index(level=1, drop=True)

# Generate a new Level 1 index (starting at 0) for each Level 0 group
new_level1 = df_temp.groupby(level=0).cumcount()

# Attach the new Level 1 index back to the DataFrame
df_final = df_temp.set_index(new_level1, append=True)

Let's Test with an Example #

Let's create a sample DataFrame to see this in action:

import numpy as np

# Create a multi-index DataFrame with NaN values
arrays = [
    ['A', 'A', 'A', 'B', 'B', 'C'],
    [0, 1, 2, 0, 1, 0]
]
df = pd.DataFrame({'value': [1, np.nan, 3, 4, 5, np.nan]}, index=arrays)

Original df:

value
A 0    1.0
  1    NaN
  2    3.0
B 0    4.0
  1    5.0
C 0    NaN

After dropna() (df_clean):

value
A 0    1.0
  2    3.0
B 0    4.0
  1    5.0

Final result (df_final):

value
A 0    1.0
  1    3.0
B 0    4.0
  1    5.0

Perfect—each Level 0 index ('A' and 'B') now has a Level 1 index starting at 0, and we kept the original Level 0 structure intact.

内容的提问来源于stack exchange，提问作者KOB