熊猫通过汇总列对双重观测进行分组

Question

I have a dataframe like this:

+----------+---------+
| username | role    |
+----------+---------+
| foo      | user    |
+----------+---------+
| foo      | analyst |
+----------+---------+
| bar      | admin   |
+----------+---------+

and I would like to remove the repetition of the users that appear twice or more by aggregating the column role in a way to obtain the following dataframe:

+----------+---------------+
| username | role          |
+----------+---------------+
| foo      | user, analyst |
+----------+---------------+
| bar      | admin         |
+----------+---------------+

So far I have tried using pivot table in this way:

table = pd.pivot_table(df, index='username', columns='role')

and also the groupby function, but this is not the right way to do it. What is the right way to deal with this?

Answer 1

What you want to do is group the rows based on username , so the groupby -function is one way to go. Usually when you use groupby you apply an aggregation function to the rest of the columns, for example sum , average , min or similair. But you can also define your own aggregation function, and use that in agg .

def merge_strings(series):
    # This function will get a series of all the values in a column. For example for foo the series will be ['user', 'analyst'].
    # We can use the built in function str.cat() fo contatenate a series of strings.

    return series.str.cat(sep=', ')

Then we simply call groupby, and tell that we want to aggregate the role -column using our custom function

df.groupby('username').agg({'role': merge_strings})

Answer 2

You can create a list or comma separate strings using the following:

df.groupby('username')['role'].agg(list).reset_index()

Output:

  username             role
0      bar          [admin]
1      foo  [user, analyst]

OR

df.groupby('username')['role'].agg(lambda x: ', '.join(x)).reset_index()

OUtput:

  username           role
0      bar          admin
1      foo  user, analyst