I have a dataset that looks like:

my_df = pd.DataFrame({"id": [1, 1, 1], "word": ["hello", "my", "friend"]})

And I would like to group by id and concatenate the word (without changing the order), the result dataset should look like:

Given:

• Dataframe is already sorted by id, words in the desired order

Constraints:

• The the word order should be maintained

CodePudding user response：

This is a .groupby operation. With pandas you can pass user defined functions, to perform aggregations. In this case you just want to .join the grouped elements with a single space.

out = my_df.groupby('id', as_index=False).agg(words=('word', ' '.join))

print(out)
   id            words
0   1  hello my friend

• as_index=False ensures that id remains a column (not the index) after the operation
• .agg(words=...) perform an aggregation on each of our groups, creating a new column titled "words"
  • ('word', ' '.join) on the column titled "word" from our grouped dataframe, apply the function ' '.join to the data. Note that you could also write this as lambda s: ' '.join(s), but I decided to go with the less verbose approach.

CodePudding user response：

If we have the following dataframe (slighyly modified from yours)

my_df = pd.DataFrame({"id": [1, 1, 1, 2, 2, 2], "word": ["hello", "my", "friend", "goodbye", "my", "friend"]})

Output

    id  word
0   1   hello
1   1   my
2   1   friend
3   2   goodbye
4   2   my
5   2   friend

To get the desired result, use

my_df.groupby('id')['word'].apply(lambda x: ' '.join(x))

Output

       id
1      hello my friend
2    goodbye my friend
Name: word, dtype: object

This groups each row by your column id, then takes each result and uses Pandas join to join them with a whitespace delimiter.

This is a groupby object, not a dataframe though; to get it back into a useable dataframe, append reset_index

    my_df.groupby('id')['word'].apply(lambda x: ' '.join(x)).reset_index()

Output

    id  word
0   1   hello my friend
1   2   goodbye my friend

If the dataframe is rather large, using apply may not be the best approach, as it performs the aggregation over every row; there may be faster methods to use. As long as your dataframe isn't tens of thousands of rows, the difference should be negligible.