Python|用列的平均值替换 NaN 值

在机器学习和数据分析中，数据可视化是最重要的步骤之一。清理和排列数据是由不同的算法完成的。有时在数据集中，我们会得到无法用于数据可视化的 NaN（不是数字）值。

为了解决这个问题，一种可能的方法是用列的平均值替换nan值。下面给出了一些解决这个问题的方法。

方法 #1：使用np.colmean和np.take

# Python code to demonstrate
# to replace nan values
# with an average of columns
  
import numpy as np
  
# Initialising numpy array
ini_array = np.array([[1.3, 2.5, 3.6, np.nan], 
                      [2.6, 3.3, np.nan, 5.5],
                      [2.1, 3.2, 5.4, 6.5]])
  
# printing initial array
print ("initial array", ini_array)
  
# column mean
col_mean = np.nanmean(ini_array, axis = 0)
  
# printing column mean
print ("columns mean", str(col_mean))
  
# find indices where nan value is present
inds = np.where(np.isnan(ini_array))
  
# replace inds with avg of column
ini_array[inds] = np.take(col_mean, inds[1])
  
# printing final array
print ("final array", ini_array)

输出：

initial array [[ 1.3  2.5  3.6  nan]
 [ 2.6  3.3  nan  5.5]
 [ 2.1  3.2  5.4  6.5]]
columns mean [ 2.   3.   4.5  6. ]

final array [[ 1.3  2.5  3.6  6. ]
 [ 2.6  3.3  4.5  5.5]
 [ 2.1  3.2  5.4  6.5]]

方法 #2：使用np.ma和np.where

# Python code to demonstrate
# to replace nan values
# with average of columns
  
import numpy as np
  
# Initialising numpy array
ini_array = np.array([[1.3, 2.5, 3.6, np.nan],
                      [2.6, 3.3, np.nan, 5.5],
                      [2.1, 3.2, 5.4, 6.5]])
  
# printing initial array
print ("initial array", ini_array)
  
# replace nan with col means
res = np.where(np.isnan(ini_array), np.ma.array(ini_array,
               mask = np.isnan(ini_array)).mean(axis = 0), ini_array)   
  
# printing final array
print ("final array", res)

输出：

initial array [[ 1.3  2.5  3.6  nan]
 [ 2.6  3.3  nan  5.5]
 [ 2.1  3.2  5.4  6.5]]
final array [[ 1.3  2.5  3.6  6. ]
 [ 2.6  3.3  4.5  5.5]
 [ 2.1  3.2  5.4  6.5]]

方法 #3：使用 Naive 和zip

# Python code to demonstrate
# to replace nan values
# with average of columns
  
import numpy as np
  
# Initialising numpy array
ini_array = np.array([[1.3, 2.5, 3.6, np.nan],
                      [2.6, 3.3, np.nan, 5.5],
                      [2.1, 3.2, 5.4, 6.5]])
  
# printing initial array
print ("initial array", ini_array)
  
# indices where values is nan in array
indices = np.where(np.isnan(ini_array))
  
# Iterating over numpy array to replace nan with values
for row, col in zip(*indices):
    ini_array[row, col] = np.mean(ini_array[
           ~np.isnan(ini_array[:, col]), col])
  
# printing final array
print ("final array", ini_array)

输出：

initial array [[ 1.3  2.5  3.6  nan]
 [ 2.6  3.3  nan  5.5]
 [ 2.1  3.2  5.4  6.5]]
final array [[ 1.3  2.5  3.6  6. ]
 [ 2.6  3.3  4.5  5.5]
 [ 2.1  3.2  5.4  6.5]]