Concatenate all columns in a pandas dataframe

Question

I have multiple pandas dataframe which may have different number of columns and the number of these columns typically vary from 50 to 100. I need to create a final column that is simply all the columns concatenated. Basically the string in the first row of the column should be the sum(concatenation) of the strings on the first row of all the columns. I wrote the loop below but I feel there might be a better more efficient way to do this. Any ideas on how to do this

num_columns = df.columns.shape[0]
col_names = df.columns.values.tolist()
df.loc[:, 'merged'] = ""
for each_col_ind in range(num_columns):
    print('Concatenating', col_names[each_col_ind])
    df.loc[:, 'merged'] = df.loc[:, 'merged'] + df[col_names[each_col_ind]]

Answer 1

Solution with sum, but output is float, so convert to int and str is necessary:

df['new'] = df.sum(axis=1).astype(int).astype(str)

Another solution with apply function join, but it the slowiest:

df['new'] = df.apply(''.join, axis=1)

Last very fast numpy solution - convert to numpy array and then 'sum':

df['new'] = df.values.sum(axis=1)

Timings:

df = pd.DataFrame({'A': ['1', '2', '3'], 'B': ['4', '5', '6'], 'C': ['7', '8', '9']})
#[30000 rows x 3 columns]
df = pd.concat([df]*10000).reset_index(drop=True)
#print (df)

cols = list('ABC')

#not_a_robot solution
In [259]: %timeit df['concat'] = pd.Series(df[cols].fillna('').values.tolist()).str.join('')
100 loops, best of 3: 17.4 ms per loop

In [260]: %timeit df['new'] = df[cols].astype(str).apply(''.join, axis=1)
1 loop, best of 3: 386 ms per loop

In [261]: %timeit df['new1'] = df[cols].values.sum(axis=1)
100 loops, best of 3: 6.5 ms per loop

In [262]: %timeit df['new2'] = df[cols].astype(str).sum(axis=1).astype(int).astype(str)
10 loops, best of 3: 68.6 ms per loop

EDIT If dtypes of some columns are not object (obviously strings) cast by DataFrame.astype:

df['new'] = df.astype(str).values.sum(axis=1)

Answer 2

df = pd.DataFrame({'A': ['1', '2', '3'], 'B': ['4', '5', '6'], 'C': ['7', '8', '9']})

df['concat'] = pd.Series(df.fillna('').values.tolist()).str.join('')

Gives us:

df
Out[6]: 
   A  B  C concat
0  1  4  7    147
1  2  5  8    258
2  3  6  9    369

To select a given set of columns:

df['concat'] = pd.Series(df[['A', 'B']].fillna('').values.tolist()).str.join('')

df
Out[8]: 
   A  B  C concat
0  1  4  7     14
1  2  5  8     25
2  3  6  9     36

However, I've noticed that approach can sometimes result in NaNs being populated where they shouldn't, so here's another way:

>>> from functools import reduce
>>> df['concat'] = df[cols].apply(lambda x: reduce(lambda a, b: a + b, x), axis=1)
>>> df
   A  B  C concat
0  1  4  7    147
1  2  5  8    258
2  3  6  9    369

Although it should be noted that this approach is a lot slower:

$ python3 -m timeit 'import pandas as pd;from functools import reduce; df=pd.DataFrame({"a": ["this", "is", "a", "string"] * 5000, "b": ["this", "is", "a", "string"] * 5000});[df[["a", "b"]].apply(lambda x: reduce(lambda a, b: a + b, x)) for _ in range(10)]'
10 loops, best of 3: 451 msec per loop

Versus

$ python3 -m timeit 'import pandas as pd;from functools import reduce; df=pd.DataFrame({"a": ["this", "is", "a", "string"] * 5000, "b": ["this", "is", "a", "string"] * 5000});[pd.Series(df[["a", "b"]].fillna("").values.tolist()).str.join(" ") for _ in range(10)]'
10 loops, best of 3: 98.5 msec per loop