16

我遇到了以下对行标题和列标题进行排序的问题。

以下是如何重现这一点：

X =pd.DataFrame(dict(x=np.random.normal(size=100), y=np.random.normal(size=100)))
A=pd.qcut(X['x'], [0,0.25,0.5,0.75,1.0]) #create a factor
B=pd.qcut(X['y'], [0,0.25,0.5,0.75,1.0]) # create another factor

g = X.groupby([A,B])['x'].mean() #do a two-way bucketing


print g 

#this gives the following and so far so good

x                 y               
[-2.315, -0.843]  [-2.58, -0.567]    -1.041167
                  (-0.567, 0.0321]   -1.722926
                  (0.0321, 0.724]    -1.245856
                  (0.724, 3.478]     -1.240876
(-0.843, -0.228]  [-2.58, -0.567]    -0.576264
                  (-0.567, 0.0321]   -0.501709
                  (0.0321, 0.724]    -0.522697
                  (0.724, 3.478]     -0.506259
(-0.228, 0.382]   [-2.58, -0.567]     0.175768
                  (-0.567, 0.0321]    0.214353
                  (0.0321, 0.724]     0.113650
                  (0.724, 3.478]     -0.013758
(0.382, 2.662]    [-2.58, -0.567]     0.983807
                  (-0.567, 0.0321]    1.214640
                  (0.0321, 0.724]     0.808608
                  (0.724, 3.478]      1.515334
Name: x, dtype: float64

#Now let's make a two way table and here is the problem:

HTML(g.unstack().to_html())

由此可见：

y                 (-0.567, 0.0321]  (0.0321, 0.724]  (0.724, 3.478]  [-2.58, -0.567]
x                                                                                   
(-0.228, 0.382]           0.214353         0.113650       -0.013758         0.175768
(-0.843, -0.228]         -0.501709        -0.522697       -0.506259        -0.576264
(0.382, 2.662]            1.214640         0.808608        1.515334         0.983807
[-2.315, -0.843]         -1.722926        -1.245856       -1.240876        -1.041167

请注意标题不再排序的方式。我想知道什么是解决这个问题的好方法，以便使交互工作变得容易。

要进一步追踪问题出在哪里，请运行以下命令：

g.unstack().columns

它给了我这个： Index([(-0.567, 0.0321], (0.0321, 0.724], (0.724, 3.478], [-2.58, -0.567]], dtype=object)

现在将其与 B.levels 进行比较：

B.levels
Index([[-2.58, -0.567], (-0.567, 0.0321], (0.0321, 0.724], (0.724, 3.478]], dtype=object)

显然Factor中原来的顺序丢失了。

现在更糟糕的是，让我们做一个多级交叉表：

g2 = X.groupby([A,B]).agg('mean')
g3 = g2.stack().unstack(-2)
HTML(g3.to_html())

它显示如下：

y                   (-0.567, 0.0321]  (0.0321, 0.724]  (0.724, 3.478]  
x                                                                       
(-0.228, 0.382]  x          0.214353         0.113650       -0.013758   
                 y         -0.293465         0.321836        1.180369   
(-0.843, -0.228] x         -0.501709        -0.522697       -0.506259   
                 y         -0.204811         0.324571        1.167005   
(0.382, 2.662]   x          1.214640         0.808608        1.515334   
                 y         -0.195446         0.161198        1.074532   
[-2.315, -0.843] x         -1.722926        -1.245856       -1.240876   
                 y         -0.392896         0.335471        1.730513

行标签和列标签都排序不正确。

谢谢。

4

1 回答 1

10

这似乎有点像黑客，但这里有：

In [11]: g_unstacked = g.unstack()

In [12]: g_unstacked
Out[12]:
y                 (-0.565, 0.12]  (0.12, 0.791]  (0.791, 2.57]  [-2.177, -0.565]
x
(-0.068, 0.625]         0.389408       0.267252       0.283344          0.258337
(-0.892, -0.068]       -0.121413      -0.471889      -0.448977         -0.462180
(0.625, 1.639]          0.987372       1.006496       0.830710          1.202158
[-3.124, -0.892]       -1.513954      -1.482813      -1.394198         -1.756679

利用unique保留 order* 的事实（从 g 的索引中获取唯一的第一项）：

In [13]: g.index.get_level_values(0).unique()
Out[13]:
array(['[-3.124, -0.892]', '(-0.892, -0.068]', '(-0.068, 0.625]',
       '(0.625, 1.639]'], dtype=object)

如您所见，这些顺序正确。

现在你可以reindex这样：

In [14]: g_unstacked.reindex(g.index.get_level_values(0).unique())
Out[14]:
y                 (-0.565, 0.12]  (0.12, 0.791]  (0.791, 2.57]  [-2.177, -0.565]
[-3.124, -0.892]       -1.513954      -1.482813      -1.394198         -1.756679
(-0.892, -0.068]       -0.121413      -0.471889      -0.448977         -0.462180
(-0.068, 0.625]         0.389408       0.267252       0.283344          0.258337
(0.625, 1.639]          0.987372       1.006496       0.830710          1.202158

现在是正确的顺序。

更新（我错过了列也没有按顺序排列）。
您可以对列使用相同的技巧（您必须链接这些操作）：

In [15]: g_unstacked.reindex_axis(g.index.get_level_values(1).unique(), axis=1)

* 这就是 Series unique 明显快于np.unique.

于 2013-06-17T21:23:08.697 回答