Merge multiple column values into one column in python pandas

I have a pandas data frame like this:

   Column1  Column2  Column3  Column4  Column5
0    a        1        2        3        4
1    a        3        4        5
2    b        6        7        8
3    c        7        7

What I want to do now is getting a new dataframe containing Column1 and a new columnA. This columnA should contain all values from columns 2 -(to) n (where n is the number of columns from Column2 to the end of the row) like this:

  Column1  ColumnA
0   a      1,2,3,4
1   a      3,4,5
2   b      6,7,8
3   c      7,7

How could I best approach this issue? Any advice would be helpful. Thanks in advance!

174239 次浏览
小开
最佳答案

You can call apply pass axis=1 to apply row-wise, then convert the dtype to str and join:

In [153]:
df['ColumnA'] = df[df.columns[1:]].apply(
lambda x: ','.join(x.dropna().astype(str)),
axis=1
)
df


Out[153]:
Column1  Column2  Column3  Column4  Column5  ColumnA
0       a        1        2        3        4  1,2,3,4
1       a        3        4        5      NaN    3,4,5
2       b        6        7        8      NaN    6,7,8
3       c        7        7      NaN      NaN      7,7

Here I call dropna to get rid of the NaN, however we need to cast again to int so we don't end up with floats as str.

小开

I propose to use .assign

df2 = df.assign(ColumnA = df.Column2.astype(str) + ', ' + \
df.Column3.astype(str) + ', ' df.Column4.astype(str) + ', ' \
df.Column4.astype(str) + ', ' df.Column5.astype(str))

it's simple, maybe long but it worked for me

小开

If you have lot of columns say - 1000 columns in dataframe and you want to merge few columns based on particular column name e.g. -Column2 in question and arbitrary no. of columns after that column (e.g. here 3 columns after 'Column2 inclusive of Column2 as OP asked).

We can get position of column using .get_loc() - as answered here

source_col_loc = df.columns.get_loc('Column2') # column position starts from 0


df['ColumnA'] = df.iloc[:,source_col_loc+1:source_col_loc+4].apply(
lambda x: ",".join(x.astype(str)), axis=1)


df


Column1  Column2  Column3  Column4  Column5  ColumnA
0       a        1        2        3        4  1,2,3,4
1       a        3        4        5      NaN    3,4,5
2       b        6        7        8      NaN    6,7,8
3       c        7        7      NaN      NaN      7,7

To remove NaN, use .dropna() or .fillna()

Hope it helps!

小开

apply() is 100X slower than agg()

Do NOT use apply, it does not scale well. Instead use df.agg(). Using apply() will take seconds, but agg() will take milliseconds (ms).

Here's an example:

import numpy as np
import pandas as pd


def createList(r1, r2):
return np.arange(r1, r2+1, 1)


sample_data = createList(1, 100_000) # a list of 100,000 values


test_df = pd.DataFrame(
[sample_data]
)


test_df.apply(lambda x: ','.join(x.astype(str))) #3.47 s ± 24.5 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)


test_df.astype(str).agg(', '.join, axis=1) #34.8 ms ± 407 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

As you can see from this sample, apply() took an average time of 3.47 seconds whereas agg() took an average time of 34.8 milliseconds. The gap in performance will become bigger as more data is added too.

*Note, I used %%timeit in jupyter notebook to get the run time for each method.


提交答案