重塑大熊猫相关矩阵

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

从上面的数据框架中,我需要将其转换为3列数据帧,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,它是相同的数据,只是以不同的方式呈现.原始数据帧中的每个列/行对简单地组合在新数据帧中的自己的行中.

不幸的是,我无法弄清楚如何将结果作为数据帧来完成.我试过了,df.stack()但结果是这样的Series.我需要它成为一个数据帧,以便我可以使用列.另一个问题df.stack()是它没有填写每一行,这里是一个小问题的样本:

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

你需要添加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