Order columns of a pandas dataframe according to the values in a row

Question

How do I order columns according to the values of the last row? In the example below, my final dataframe should have columns in the following order: 'ddd' 'aaa' 'ppp' 'fff'.

>>> df = DataFrame(np.random.randn(10, 4), columns=['ddd', 'fff', 'aaa', 'ppp'])
>>> df
        ddd       fff       aaa       ppp
0 -0.177438  0.102561 -1.318710  1.321252
1  0.980348  0.786721  0.374506 -1.411019
2  0.405112  0.514216  1.761983 -0.529482
3  1.659710 -1.017048 -0.737615 -0.388145
4 -0.472223  1.407655 -0.129119 -0.912974
5  1.221324 -0.656599  0.563152 -0.900710
6 -1.816420 -2.898094 -0.232047 -0.648904
7  2.793261  0.568760 -0.850100  0.654704
8 -2.180891  2.054178 -1.050897 -1.461458
9 -1.123756  1.245987 -0.239863  0.359759

Answer 1

[updated to simplify]

tl;dr:

In [29]: new_columns = df.columns[df.ix[df.last_valid_index()].argsort()]

In [30]: df[new_columns]
Out[30]: 
        aaa       ppp       fff       ddd
0  0.328281  0.375458  1.188905  0.503059
1  0.305457  0.186163  0.077681 -0.543215
2  0.684265  0.681724  0.210636 -0.532685
3 -1.134292  1.832272  0.067946  0.250131
4 -0.834393  0.010211  0.649963 -0.551448
5 -1.032405 -0.749949  0.442398  1.274599

---

Some explanation follows. First, build the DataFrame :

In [24]: df = pd.DataFrame(np.random.randn(6, 4), columns=['ddd', 'fff', 'aaa', 'ppp'])

In [25]: df
Out[25]: 
        ddd       fff       aaa       ppp
0  0.503059  1.188905  0.328281  0.375458
1 -0.543215  0.077681  0.305457  0.186163
2 -0.532685  0.210636  0.684265  0.681724
3  0.250131  0.067946 -1.134292  1.832272
4 -0.551448  0.649963 -0.834393  0.010211
5  1.274599  0.442398 -1.032405 -0.749949

Get the last row:

In [26]: last_row = df.ix[df.last_valid_index()]

Get the indices that would sort it:

In [27]: last_row.argsort()
Out[27]: 
ddd    2
fff    3
aaa    1
ppp    0
Name: 5, Dtype: int32

Use this to index df :

In [28]: df[last_row.argsort()]
Out[28]: 
        aaa       ppp       fff       ddd
0  0.328281  0.375458  1.188905  0.503059
1  0.305457  0.186163  0.077681 -0.543215
2  0.684265  0.681724  0.210636 -0.532685
3 -1.134292  1.832272  0.067946  0.250131
4 -0.834393  0.010211  0.649963 -0.551448
5 -1.032405 -0.749949  0.442398  1.274599

Profit!

Answer 2

The sort_values method does this directly when given axis=1 argument.

sorted_df = df.sort_values(df.last_valid_index(), axis=1)

---

So, it is no longer necessary to transpose the dataframe to sort by a row. Also, the sort method is now deprecated.

Answer 3

I would use transpose and the sort method (which works on columns):

df = pd.DataFrame(np.random.randn(10, 4), columns=['ddd', 'fff', 'aaa', 'ppp'])
last_row_name = df.index[-1]
sorted_df = df.T.sort(columns=last_row_name).T

You might suffer a performance hit but it is quick and easy.

Answer 4

df=df[df.iloc[-1,:].sort_values().index]

这有效