[FEA] DataFrame swapaxes

[beckernick]

For pandas API compatibility, we can implement DataFrame.swapaxes. This appears to either return a transpose (swapping index and column axes) or the original frame. In both cases, the data is copied by default (but can be parameterized to return a view / the original frame instead).

Series.swapaxes is likely going to be deprecated in the future based on pandas-dev/pandas#18262 and the method itself is generally no longer useful since pandas removed Panels. There may be some numpy compatibility reasons it still exists, based on pandas-dev/pandas#26654 . If we eventually run into similar needs with CuPy, we can cross that bridge when we come to it.

import pandas as pd
​
df = pd.DataFrame({
    "a": [0,1,None,3],
    "b": [3,9,20,1]
})
​
print(df)
print(df.swapaxes(0,0, copy=False))
print(df.swapaxes("index", "columns"))
print(df.swapaxes(1,0))
print(df.swapaxes(0,1))
     a   b
0  0.0   3
1  1.0   9
2  NaN  20
3  3.0   1
     a   b
0  0.0   3
1  1.0   9
2  NaN  20
3  3.0   1
     0    1     2    3
a  0.0  1.0   NaN  3.0
b  3.0  9.0  20.0  1.0
     0    1     2    3
a  0.0  1.0   NaN  3.0
b  3.0  9.0  20.0  1.0
     0    1     2    3
a  0.0  1.0   NaN  3.0
b  3.0  9.0  20.0  1.0

[vyasr]

DataFrame.swapaxes was deprecated in pandas 2.1 (this issue) so it's not worth our time to implement it now.