PERF/API: alias len/size in groupby.transform

[jreback]

alias len/size to groupby.transform speedup table (as these are already implemented as size,`count`` in cython, just a matter of transforming them

see here: http://stackoverflow.com/questions/24438273/aggregation-on-pandas-datetime-series-only-returns-as-datetime-series/24438366#24438366
(len is suspect for .agg as well)

so

g = df.groupby(...)

g.transform(len)
g.transform('size')
g.transform('len')

will do the same

[dsm054]

Just to make sure, you're not proposing removing the distinction between len and count?

[jreback]

hmm

count would be number of non Nan's in group
but I think that is the same as the group size
yes? (I suppose u could have a count of non-nan rows but simpler to just dropna before grouping) - this is a group size (but it transforms)

so I gues len==size==count (and want to accept built in len function)

[dsm054]

If memory serves Andy raised this point in the past:

>>> df = pd.DataFrame({"a": [1,1,2,2,3], "b": [1,1,np.nan,2,3]})
>>> df.groupby("a").transform(len)
   b
0  2
1  2
2  2
3  2
4  1
>>> df.groupby("a").transform("count")
   b
0  2
1  2
2  1
3  1
4  1

[jreback]

ok yah that's right
ok then just len==size (as strings and len as a builtin)

[jreback]

side issue: https://groups.google.com/forum/#!topic/pydata/xinuPFPVakY

I guess .agg(min) is NOT calling .min()

[jreback]

going make a new issue about that last one