当两个numpy数组中有一个具有NAN值时，如何计算它们之间的相关系数？

Question

下面的数组是较长系列的摘录

我试过这个：

np.corrcoef(A1, A2)

得到这个：

array([[ 1., nan],
       [nan, nan]])

阵列A1:

array([118.76217 , 119.29147 , 119.737   , 120.0961  , 120.66373 ,
       121.325195, 121.86492 , 122.27655 , 122.59397 , 122.97091 ,
       123.84733 , 125.23529 , 126.442024, 127.58224 , 128.59303 ,
       129.46916 , 130.55403 , 132.379   , 134.57579 , 136.9152  ,
       139.08174 , 140.94403 , 142.54588 , 144.08707 , 145.62819 ,
       147.26051 , 148.82619 , 150.28763 , 152.11078 , 153.83958 ,
       155.80728 , 158.07167 , 160.01866 , 162.40714 , 165.73    ,
       168.6646  , 171.11201 , 173.11388 , 174.95331 , 177.12701 ,
       179.31892 , 181.48216 , 183.3753  , 185.30406 , 187.08716 ,
       189.45274 , 191.74364 , 193.79718 , 196.03215 , 198.83864 ,
       202.0072  , 204.65758 , 206.76361 , 208.48698 , 210.4281  ,
       212.42377 , 214.2105  , 215.89319 , 218.44202 , 221.37914 ,
       224.42348 , 226.92468 , 228.8517  ], dtype=float32)

数组A2:

array([     nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan, 187.253 , 179.628 , 169.1065, 159.6525,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan, 187.253 , 179.1705, 168.649 ,
       159.5   ,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan,      nan,      nan,      nan,
            nan,      nan,      nan])

Answer 1

一种选择是使用屏蔽数组：

import numpy as np
import numpy.ma as ma

A1 = np.array([118.76217 , 119.29147 , 119.737   , 120.0961  , 120.66373 ,
               121.325195, 121.86492 , 122.27655 , 122.59397 , 122.97091 ,
               123.84733 , 125.23529 , 126.442024, 127.58224 , 128.59303 ,
               129.46916 , 130.55403 , 132.379   , 134.57579 , 136.9152  ,
               139.08174 , 140.94403 , 142.54588 , 144.08707 , 145.62819 ,
               147.26051 , 148.82619 , 150.28763 , 152.11078 , 153.83958 ,
               155.80728 , 158.07167 , 160.01866 , 162.40714 , 165.73    ,
               168.6646  , 171.11201 , 173.11388 , 174.95331 , 177.12701 ,
               179.31892 , 181.48216 , 183.3753  , 185.30406 , 187.08716 ,
               189.45274 , 191.74364 , 193.79718 , 196.03215 , 198.83864 ,
               202.0072  , 204.65758 , 206.76361 , 208.48698 , 210.4281  ,
               212.42377 , 214.2105  , 215.89319 , 218.44202 , 221.37914 ,
               224.42348 , 226.92468 , 228.8517  ])

A2 = np.array([np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan, 187.253,  179.628, 169.1065, 159.6525,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,  187.253, 179.1705,  168.649,
                159.5, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan,   np.nan,   np.nan,   np.nan,
               np.nan, np.nan,  np.nan])

print(ma.corrcoef(ma.masked_invalid(A1), ma.masked_invalid(A2)))

# Prints:
#[[1.0 -0.07135569546454648]
# [-0.07135569546454648 1.0]]

或者，可以将数组存储在pandas数据帧中，并使用df.corr()方法，该方法对Nan友好

Answer 2

这将在使用屏蔽阵列numpy模块时起作用：

import numpy as np
import numpy.ma as ma

A = np.asarray([1.12, 2.34, 3.33])
B = np.asarray([1.12, float('Inf') , 3.33])


print(ma.corrcoef(ma.masked_invalid(A), ma.masked_invalid(B)))

输出：

[[1.0 1.0]                                                                                                                                
 [1.0 1.0]]