PERF: DataFrame groupby with fast transform

[jreback]

from SO

import pandas as pd
import numpy as np

df = pd.DataFrame({'group': np.repeat(np.arange(1000), 10),
                   'B': np.nan,
                   'C': np.nan})

df.ix[4::10, 'B':'C'] = 5 # every 4th row of a group is non-null

df.groupby('group').transform('first')

This is then iterating over groups. Last I can see this was changed is: here. My recollection is that this was ONLY supposed to hit in a special case, and the general case is simply a repeat based on the indices.

This seems to be hitting in all cases makes transform back to super SLOW.

[jreback]

cc @ajcr
cc @chris-b1
cc @evanpw

[jreback]

So we have a fast path for Series transforms, but not for DataFrame transforms.

In [17]: result1 = g.transform('first')

In [18]: result2 = pd.concat([g.B.transform('first'), g.C.transform('first')], keys=['B','C'], axis=1)

In [19]: result1.equals(result2)
Out[19]: True

In [20]: %timeit g.transform('first')
10 loops, best of 3: 170 ms per loop

In [21]: %timeit pd.concat([g.B.transform('first'), g.C.transform('first')], keys=['B','C'], axis=1)
1000 loops, best of 3: 2 ms per loop