如何用groupby后单列nunique值筛选ID？为何单列操作报错？

Great question! Let's break down exactly why this behavior happens and how to fix the single-column case.

First, the core difference boils down to the type of object pandas returns in each scenario:

• Single-column groupby.nunique: Returns a pandas Series (1-dimensional data structure). The index of this Series is your id values, and the values are the nunique() counts for random_column. While the Series has a name of random_column, you can't access it like a DataFrame column with .random_column—Series don't support that attribute-style column access.
• Multi-column groupby.nunique: Returns a pandas DataFrame (2-dimensional structure). Each original column you included becomes a named column in the result, so you can use .random_column to reference that specific column for filtering.

Let's Walk Through Concrete Examples #

Say we have this sample DataFrame:

import pandas as pd
df = pd.DataFrame({
    'id': [1,1,2,3,3,3],
    'random_column': ['a','b','a','a','b','c'],
    'one_more_column': [1,1,2,2,2,3]
})

Single-Column Scenario #

Running:

temp = df.groupby('id')['random_column'].nunique()

Gives us a Series output:

id
1    2
2    1
3    3
Name: random_column, dtype: int64

Trying temp.random_column > 1 throws an AttributeError because Series don't have column attributes like DataFrames. Instead, filter directly on the Series itself:

# Correct way to filter the Series and get the IDs
filtered_ids = temp[temp > 1].index.tolist()

This gives you the list of IDs where random_column has more than 1 unique value: [1, 3].

If you prefer to use DataFrame-style column access, convert the Series to a DataFrame with to_frame():

temp_df = df.groupby('id')['random_column'].nunique().to_frame()
filtered_ids = temp_df[temp_df.random_column > 1].index.tolist()

Now this works because temp_df is a full DataFrame with a column named random_column.

Multi-Column Scenario #

Running:

temp = df.groupby('id')['random_column', 'one_more_column'].nunique()

Gives us a DataFrame output:

random_column  one_more_column
id
1              2                1
2              1                1
3              3                2

Here, temp.random_column correctly references the first column, so temp[temp.random_column > 1] works exactly as you expect.

Key Takeaway #

• For single-column groupby results (Series), use direct boolean indexing (temp > 1) to filter values.
• If you want to stick with DataFrame-style column access, convert the Series to a DataFrame using to_frame().

内容的提问来源于stack exchange，提问作者loadbox