Ability to Sort DataFrame by Transformed Values

[cancan101]

Let's say I have a DataFrame of stock returns. It would be great to be able to sort easily the DataFrame by the abs value of the returns. Currently I do this by inserting a new column and then sorting on the new column and than slicing on the original columns. This is ungainly:

ret["returns_abs"] = ret.returns.abs()
ret.sort("returns_abs", ascending=False)[["returns", "key"]]

It would be great if I could either pass a Series or a function to the sort method:

ret.sort(ret.returns.abs(), ascending=False)

or

ret.sort(lambda x: abs(x.ix["returns"], ascending=False)

[jreback]

just reindex

[cancan101]

I assume that I am doing something wrong here:

df = pd.DataFrame(np.random.rand(13, 3), columns=list('abc'))

In [148]:

print df.reindex(df.a.abs())

           a   b   c
a                   
0.074637 NaN NaN NaN
0.348153 NaN NaN NaN
0.751475 NaN NaN NaN
0.062545 NaN NaN NaN
0.117850 NaN NaN NaN
0.296180 NaN NaN NaN
0.729935 NaN NaN NaN
0.814809 NaN NaN NaN
0.935224 NaN NaN NaN
0.960110 NaN NaN NaN
0.121138 NaN NaN NaN
0.103864 NaN NaN NaN
0.808862 NaN NaN NaN

[13 rows x 3 columns]

[jreback]

In [19]: df = pd.DataFrame(np.random.rand(5, 3), columns=list('abc'))-.5

In [20]: df
Out[20]: 
          a         b         c
0  0.086302  0.280969 -0.472230
1 -0.232537 -0.159322  0.079688
2 -0.467258  0.380615 -0.239272
3 -0.334849 -0.011547  0.499909
4 -0.060898  0.180957 -0.002005

[5 rows x 3 columns]

In [21]: df.reindex(index=df.a.abs().order().index)
Out[21]: 
          a         b         c
4 -0.060898  0.180957 -0.002005
0  0.086302  0.280969 -0.472230
1 -0.232537 -0.159322  0.079688
3 -0.334849 -0.011547  0.499909
2 -0.467258  0.380615 -0.239272

[5 rows x 3 columns]

[jreback]

see also #3942

[jreback]

this issue is a dupe of that