熔化 Pandas Dataframe 的上三角矩阵

Melt the Upper Triangular Matrix of a Pandas Dataframe

给定一个正方形 pandas DataFrame,其形式如下:

   a  b  c
a  1 .5 .3
b .5  1 .4
c .3 .4  1

如何将upper triangle熔化得到如下形式的矩阵

 Row     Column    Value
  a        a       1
  a        b       .5 
  a        c       .3
  b        b       1
  b        c       .4
  c        c       1 

#Note the combination a,b is only listed once.  There is no b,a listing     

我对惯用的 pandas 解决方案更感兴趣,自定义索引器很容易手写...

提前感谢您的考虑和回复。

首先,我通过 where and numpy.triu and then stack, reset_indexdf 的较低值转换为 NaN 并设置列名称:

import numpy as np

print df
     a    b    c
a  1.0  0.5  0.3
b  0.5  1.0  0.4
c  0.3  0.4  1.0

print np.triu(np.ones(df.shape)).astype(np.bool)
[[ True  True  True]
 [False  True  True]
 [False False  True]]

df = df.where(np.triu(np.ones(df.shape)).astype(np.bool))
print df
    a    b    c
a   1  0.5  0.3
b NaN  1.0  0.4
c NaN  NaN  1.0

df = df.stack().reset_index()
df.columns = ['Row','Column','Value']
print df

  Row Column  Value
0   a      a    1.0
1   a      b    0.5
2   a      c    0.3
3   b      b    1.0
4   b      c    0.4
5   c      c    1.0

从@jezrael 的解决方案构建,布尔索引将是一种更明确的方法:

import numpy
from pandas import DataFrame

df = DataFrame({'a':[1,.5,.3],'b':[.5,1,.4],'c':[.3,.4,1]},index=list('abc'))
print df,'\n'
keep = np.triu(np.ones(df.shape)).astype('bool').reshape(df.size)
print df.stack()[keep]

输出:

     a    b    c
a  1.0  0.5  0.3
b  0.5  1.0  0.4
c  0.3  0.4  1.0 

a  a    1.0
   b    0.5
   c    0.3
b  b    1.0
   c    0.4
c  c    1.0
dtype: float64

同样建立在@jezrael 的解决方案之上,这是一个版本,添加了一个函数来执行逆运算(从 xy 到矩阵),在我的情况下有用的协方差/相关矩阵。

def matrix_to_xy(df, columns=None, reset_index=False):
    bool_index = np.triu(np.ones(df.shape)).astype(bool)
    xy = (
        df.where(bool_index).stack().reset_index()
        if reset_index
        else df.where(bool_index).stack()
    )
    if reset_index:
        xy.columns = columns or ["row", "col", "val"]
    return xy


def xy_to_matrix(xy):
    df = xy.pivot(*xy.columns).fillna(0)
    df_vals = df.to_numpy()
    df = pd.DataFrame(
        np.triu(df_vals, 1) + df_vals.T, index=df.index, columns=df.index
    )
    return df
df = pd.DataFrame(
    {"a": [1, 0.5, 0.3], "b": [0.5, 1, 0.4], "c": [0.3, 0.4, 1]},
    index=list("abc"),
)
print(df)
xy = matrix_to_xy(df, reset_index=True)
print(xy)
mx = xy_to_matrix(xy)
print(mx)

输出:

     a    b    c
a  1.0  0.5  0.3
b  0.5  1.0  0.4
c  0.3  0.4  1.0

  row col  val
0   a   a  1.0
1   a   b  0.5
2   a   c  0.3
3   b   b  1.0
4   b   c  0.4
5   c   c  1.0

row    a    b    c
row
a    1.0  0.5  0.3
b    0.5  1.0  0.4
c    0.3  0.4  1.0