Is there a reset_index for columns or a way to move column headers to an inner index leaving their index positions as the outer index?

Question

Sample DataFrame:

import numpy as np
df = pd.DataFrame(np.random.randint(0, 10, size=(10, 4)), columns=list('ABCD'))

Is there a way to reset index for columns? or to easily insert a row with column index position values? I'd prefer the index positions to be the outer most index and be left with the column headers as the inner most index.

Answer 1

a.1) Drop column names

df.columns = pd.RangeIndex(df.columns.size)
df

Output:

    0   1   2   3
#---------------#
0   0   1   3   3
1   2   2   0   2
2   2   1   3   1
3   2   1   0   0

a.2) Drop column names (one-liner)
Could have performance issues and side effects, see discussion below.

df.T.reset_index(drop=True).T 

Output:

    0   1   2   3
#---------------#
0   0   1   3   3
1   2   2   0   2
2   2   1   3   1
3   2   1   0   0

b.1) Move column names into a row (one-liner)
Same issues, see discussion below.

df.T.reset_index().T

Output:

        0   1   2   3
#-------------------#
index   A   B   C   D
   0    0   1   3   3
   1    2   2   0   2
   2    2   1   3   1
   3    2   1   0   0

b.2) Move column names into a row
Effective way.

 #heterogeneous DataFrame creation
df = pd.DataFrame(np.random.randint(0,4,size=(4, 3)), columns=list('789')).join(
     pd.DataFrame(list('bcde'),columns=['A']))
df.index.name = '4'

#save column as row then reindex column names
df = df.append(pd.Series( df.columns,name = df.index.name,index= df.columns ), )
df.columns = pd.RangeIndex(df.columns.size)
print (df)
print(df.info())

Output: NB you will need extra effort to prevent upcasing of all data

   0  1  2  3
#-----------#
4            
0  2  3  2  b
1  1  0  2  c
2  3  1  3  d
3  3  3  2  e
4  7  8  9  A

<class 'pandas.core.frame.DataFrame'>
Index: 5 entries, 0 to 4
Data columns (total 4 columns):
0    5 non-null object
1    5 non-null object
2    5 non-null object
3    5 non-null object
dtypes: object(4)

c) Add secondary column index (one-liner)
Could have performance issues and side effects, see discussion below.

df.T.set_index(pd.RangeIndex(df.columns.size),append=True).T

Output:

    A   B   C   D
    0   1   2   3
#---------------#
0   0   1   3   3
1   2   2   0   2
2   2   1   3   1
3   2   1   0   0

---

One line approach criticism

Performance issues:
For huge datasets could be unacceptable costs of double T , but on simple cases one line that returns copy of DataFrame maybe useful. See test results

In [294]: for i in range (3,7):
     ...:     df = pd.DataFrame(np.random.randint(0,9,size=(10**i, 10**3)))
     ...:     print ('shape:',df.shape)
     ...:     %timeit df.T.reset_index(drop=True)
     ...: 
shape: (1000, 1000)
100 loops, best of 3: 3.2 ms per loop
shape: (10000, 1000)
10 loops, best of 3: 29.3 ms per loop
shape: (100000, 1000)
1 loop, best of 3: 546 ms per loop
shape: (1000000, 1000)
1 loop, best of 3: 9.9 s per loop

In [295]: %timeit df.columns = pd.RangeIndex(df.columns.size)
The slowest run took 28.60 times longer than the fastest. This could mean that an intermediate result is being cached.
100000 loops, best of 3: 7.74 µs per loop
  

Side effect (upcasting):
Heterogeneous DataFrames will be up-casted

In [352]: df = pd.DataFrame(np.random.randint(0,4,size=(4, 3)), columns=list('789')).join(
     ...:          pd.DataFrame(list('bcde'),columns=['A']))

In [353]: df.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4 entries, 0 to 3
Data columns (total 4 columns):
7    4 non-null int64
8    4 non-null int64
9    4 non-null int64
A    4 non-null object
dtypes: int64(3), object(1)
memory usage: 208.0+ bytes

.T.T upcasting

In [354]: df.T.T.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4 entries, 0 to 3
Data columns (total 4 columns):
7    4 non-null object
8    4 non-null object
9    4 non-null object
A    4 non-null object
dtypes: object(4)
memory usage: 208.0+ bytes

---

Answer 2

I think you can use numpy.arange or range:

np.random.seed(10)
df = pd.DataFrame(np.random.randint(0,10,size=(10, 4)), columns=list('ABCD'))

df.columns = np.arange(len(df.columns))
#alternatively
#df.columns = range(len(df.columns))
print (df)
   0  1  2  3
0  9  4  0  1
1  9  0  1  8
2  9  0  8  6
3  4  3  0  4
4  6  8  1  8
5  4  1  3  6
6  5  3  9  6
7  9  1  9  4
8  2  6  7  8
9  8  9  2  0

But lost column values.

If need MultiIndex without names:

df.columns = [np.arange(len(df.columns)), df.columns]
print (df)
   0  1  2  3
   A  B  C  D
0  9  4  0  1
1  9  0  1  8
2  9  0  8  6
3  4  3  0  4
4  6  8  1  8
5  4  1  3  6
6  5  3  9  6
7  9  1  9  4
8  2  6  7  8
9  8  9  2  0

and for names use MultiIndex.from_arrays:

names = ['a','b']
df.columns = pd.MultiIndex.from_arrays([np.arange(len(df.columns)), df.columns], names=names)
print (df)
a  0  1  2  3
b  A  B  C  D
0  9  4  0  1
1  9  0  1  8
2  9  0  8  6
3  4  3  0  4
4  6  8  1  8
5  4  1  3  6
6  5  3  9  6
7  9  1  9  4
8  2  6  7  8
9  8  9  2  0