Pandas: Sorting columns by their mean value

Question

I have a dataframe in Pandas, I would like to sort its columns (i.e. get a new dataframe, or a view) according to the mean value of its columns (or e.g. by their std value). The documentation talks about sorting by label or value, but I could not find anything on custom sorting methods.

How can I do this?

Answer 1

You can use the mean DataFrame method and the Series sort_values method:

In [11]: df = pd.DataFrame(np.random.randn(4,4), columns=list('ABCD'))  In [12]: df Out[12]:           A         B         C         D 0  0.933069  1.432486  0.288637 -1.867853 1 -0.455952 -0.725268  0.339908  1.318175 2 -0.894331  0.573868  1.116137  0.508845 3  0.661572  0.819360 -0.527327 -0.925478  In [13]: df.mean() Out[13]: A    0.061089 B    0.525112 C    0.304339 D   -0.241578 dtype: float64  In [14]: df.mean().sort_values() Out[14]: D   -0.241578 A    0.061089 C    0.304339 B    0.525112 dtype: float64 

Then you can reorder the columns using reindex:

In [15]: df.reindex(df.mean().sort_values().index, axis=1) Out[15]:           D         A         C         B 0 -1.867853  0.933069  0.288637  1.432486 1  1.318175 -0.455952  0.339908 -0.725268 2  0.508845 -0.894331  1.116137  0.573868 3 -0.925478  0.661572 -0.527327  0.819360 

---

Note: In earlier versions of pandas, sort_values used to be order, but order was deprecated as part of 0.17 so to be more consistent with the other sorting methods. Also, in earlier versions, one had to use reindex_axis rather than reindex.

Answer 2

You can use assign to create a variable, use it to sort values and drop it in the same line of code.

df = pd.DataFrame(np.random.randn(4,4), columns=list('ABCD')) df.assign(m=df.mean(axis=1)).sort_values('m').drop('m', axis=1)