重塑 pandas 相关矩阵

Question

我有以下使用 pandas 创建的相关矩阵：df.corr()

symbol       aaa       bbb       ccc       ddd       eee
symbol                                                  
aaa     1.000000  0.346099  0.131874 -0.150910  0.177589
bbb     0.346099  1.000000  0.177308 -0.384893  0.301150
ccc     0.131874  0.177308  1.000000 -0.176995  0.258812
ddd    -0.150910 -0.384893 -0.176995  1.000000 -0.310137
eee     0.177589  0.301150  0.258812 -0.310137  1.000000

根据上面的数据框，我需要将其转换为 3 列数据框，如下所示：

aaa     aaa       1.000000
aaa     bbb       0.346099
aaa     ccc       0.131874
aaa     ddd      -0.150910
aaa     eee       0.177589
bbb     aaa       0.346099
bbb     bbb       1.000000
bbb     ccc       0.177308
bbb     ddd      -0.384893
bbb     eee       0.301150
ccc     aaa       0.131874
ccc     bbb       0.177308
ccc     ccc       1.000000
ccc     ddd      -0.176995
ccc     eee       0.258812
ddd     aaa      -0.150910
ddd     bbb      -0.384893
ddd     ccc      -0.176995
ddd     ddd       1.000000
ddd     eee      -0.310137
eee     aaa       0.177589
eee     bbb       0.301150
eee     ccc       0.258812
eee     ddd      -0.310137
eee     eee       1.000000

如图所示，同样是数据，只是呈现方式不同。原始数据帧中的每个 column/row 对都简单地分组到新数据帧中它自己的行中。

不幸的是，我无法弄清楚如何在结果是数据帧的情况下完成这项工作。我试过 df.stack() 但结果是 Series。我需要它是一个数据框，以便我可以使用这些列。 df.stack() 的另一个问题是它不会填写每一行，这里是该问题的一个小示例：

aaa     aaa       1.000000
        bbb       0.346099
        ccc       0.131874
        ddd      -0.150910
        eee       0.177589
bbb     aaa       0.346099
        bbb       1.000000
        ccc       0.177308
        ddd      -0.384893
        eee       0.301150
etc...

Answer 1

您需要添加 reset_index:

#reset columns and index names 
df = df.rename_axis(None).rename_axis(None, axis=1)

#if pandas version below 0.18.0
#df.columns.name = None
#df.index.name = None

print (df)
          aaa       bbb       ccc       ddd       eee
aaa  1.000000  0.346099  0.131874 -0.150910  0.177589
bbb  0.346099  1.000000  0.177308 -0.384893  0.301150
ccc  0.131874  0.177308  1.000000 -0.176995  0.258812
ddd -0.150910 -0.384893 -0.176995  1.000000 -0.310137
eee  0.177589  0.301150  0.258812 -0.310137  1.000000

df1 = df.stack().reset_index()
#set column names
df1.columns = ['a','b','c']
print (df1)
      a    b         c
0   aaa  aaa  1.000000
1   aaa  bbb  0.346099
2   aaa  ccc  0.131874
3   aaa  ddd -0.150910
4   aaa  eee  0.177589
5   bbb  aaa  0.346099
6   bbb  bbb  1.000000
7   bbb  ccc  0.177308
8   bbb  ddd -0.384893
9   bbb  eee  0.301150
10  ccc  aaa  0.131874
11  ccc  bbb  0.177308
12  ccc  ccc  1.000000
13  ccc  ddd -0.176995
14  ccc  eee  0.258812
15  ddd  aaa -0.150910
16  ddd  bbb -0.384893
17  ddd  ccc -0.176995
18  ddd  ddd  1.000000
19  ddd  eee -0.310137
20  eee  aaa  0.177589
21  eee  bbb  0.301150
22  eee  ccc  0.258812
23  eee  ddd -0.310137
24  eee  eee  1.000000

Answer 2

使用下面的代码 (a) 重塑相关矩阵，(b) 删除重复行（例如，{aaa, bbb} 和 {bbb, aaa})，以及 (c) 删除前两列中包含相同变量的行（例如，{aaa, aaa}）：

# calculate the correlation matrix and reshape
df_corr = df.corr().stack().reset_index()

# rename the columns
df_corr.columns = ['FEATURE_1', 'FEATURE_2', 'CORRELATION']

# create a mask to identify rows with duplicate features as mentioned above
mask_dups = (df_corr[['FEATURE_1', 'FEATURE_2']].apply(frozenset, axis=1).duplicated()) | (df_corr['FEATURE_1']==df_corr['FEATURE_2']) 

# apply the mask to clean the correlation dataframe
df_corr = df_corr[~mask_dups]

这将生成如下输出：

    FEATURE_1  FEATURE_2  CORRELATION
0         aaa        bbb     0.346099
1         aaa        ccc     0.131874
2         aaa        ddd    -0.150910
3         aaa        eee     0.177589
4         bbb        ccc     0.177308
5         bbb        ddd    -0.384893
6         bbb        eee     0.301150
7         ccc        ddd    -0.176995
8         ccc        eee     0.258812
9         ddd        eee    -0.310137

Answer 3

单线解决方案：

df.corr().stack().rename_axis(('a', 'b')).reset_index(name='value')

重塑 pandas 相关矩阵

Reshaping a pandas correlation matrix

python

matrix

reshape

dataframe

pandas