Combining columns of row data frames into a single column in Pandas

Question

I have columns in a data frame (imported from CSV) containing text like this.

"New york", "Atlanta", "Mumbai"
"Beijing", "Paris", "Budapest"
"Brussels", "Oslo", "Singapore"

I want to collapse / merge all columns into one column, e.g.

New york Atlanta
Beijing Paris Budapest
Brussels Oslo Singapore

How to do it in pandas?

Answer 1

Suppose you have DataFrame:

>>> df
          0        1          2
0  New york  Atlanta     Mumbai
1   Beijing    Paris   Budapest
2  Brussels     Oslo  Singapore

Then a simple use of the method pd.DataFrame.applywill work beautifully:

>>> df.apply(" ".join, axis=1)
0    New york Atlanta Mumbai
1     Beijing Paris Budapest
2    Brussels Oslo Singapore
dtype: object

Note. I need to pass axis=1so that it is applied column by column, not row by row. I.e:

>>> df.apply(" ".join, axis=0)
0    New york Beijing Brussels
1           Atlanta Paris Oslo
2    Mumbai Budapest Singapore
dtype: object

Answer 2

Faster (but ugly) version with .cat:

df[0].str.cat(df.ix[:, 1:].T.values, sep=' ')

0    New york Atlanta Mumbai
1     Beijing Paris Budapest
2    Brussels Oslo Singapore
Name: 0, dtype: object

In a larger (10kx5) DataFrame:

%timeit df.apply(" ".join, axis=1)
10 loops, best of 3: 112 ms per loop

%timeit df[0].str.cat(df.ix[:, 1:].T.values, sep=' ')
100 loops, best of 3: 4.48 ms per loop

Answer 3

:

def pir(df):
    df = df.copy()
    df.insert(2, 's', ' ', 1)
    df.insert(1, 's', ' ', 1)
    return df.sum(1)

def pir2(df):
    df = df.copy()
    return pd.MultiIndex.from_arrays(df.values.T).to_series().str.join(' ').reset_index(drop=True)

def pir3(df):
    a = df.values[:, 0].copy()
    for j in range(1, df.shape[1]):
        a += ' ' + df.values[:, j]
    return pd.Series(a)

---

Timing

pir3 df

pir3 df 30 000

Answer 4

:

In [160]: df1.add([' '] * (df1.columns.size - 1) + ['']).sum(axis=1)
Out[160]:
0    New york Atlanta Mumbai
1     Beijing Paris Budapest
2    Brussels Oslo Singapore
dtype: object

:

In [162]: [' '] * (df.columns.size - 1) + ['']
Out[162]: [' ', ' ', '']

300K DF:

In [68]: df = pd.concat([df] * 10**5, ignore_index=True)

In [69]: df.shape
Out[69]: (300000, 3)

In [76]: %timeit df.apply(" ".join, axis=1)
1 loop, best of 3: 5.8 s per loop

In [77]: %timeit df[0].str.cat(df.ix[:, 1:].T.values, sep=' ')
10 loops, best of 3: 138 ms per loop

In [79]: %timeit pir(df)
1 loop, best of 3: 499 ms per loop

In [80]: %timeit pir2(df)
10 loops, best of 3: 174 ms per loop

In [81]: %timeit pir3(df)
10 loops, best of 3: 115 ms per loop

In [159]: %timeit df.add([' '] * (df.columns.size - 1) + ['']).sum(axis=1)
1 loop, best of 3: 478 ms per loop

Conclusion: current winner @piRSquared pir3 ()

Answer 5

If you prefer something more explicit ...

Starting with a dataframe df that looks like this:

>>> df
          A         B          C
0  New york   Beijing   Brussels
1   Atlanta     Paris       Oslo
2    Mumbai  Budapest  Singapore

You can create a new column as follows:

df['result'] = df['A'] + ' ' + df['B'] + ' ' + df['C']

In this case, the result is stored in the "result" column of the original DataFrame:

          A         B          C                     result
0  New york   Beijing   Brussels  New york Beijing Brussels
1   Atlanta     Paris       Oslo         Atlanta Paris Oslo
2    Mumbai  Budapest  Singapore  Mumbai Budapest Singapore