Question

我有类似的情况 Pandas Group Weighted Average of Multiple Columns，但其中一列的某些值有时为NaN。

也就是说，我正在执行以下操作：

import pandas as pd
import numpy as np

df=pd.DataFrame({'category':['a','a','b','b'],
 'var1':np.random.randint(0,100,4),
 'var2':np.random.randint(0,100,4),
 'weights':np.random.randint(0,10,4)})
df.loc[1,'var1']=np.nan
df


      category  var1  var2  weights
0        a      74.0    99        9
1        a       NaN     8        4
2        b      13.0    86        2
3        b      49.0    38        7

def weighted(x, cols, w="weights"):
    # Following fails when NaNs might be present:
    #return pd.Series(np.average(x[cols], weights=x[w], axis=0), cols)
    return pd.Series([np.nan if x.dropna(subset=[c]).empty else np.average(x.dropna(subset=[c])[c], weights =x.dropna(subset=[c])[w] ) for c in cols], cols)

df.groupby('category').apply(weighted, ['var1', 'var2'])


          var1       var2
category                 
a         74.0  57.846154
b         23.0   8.000000

我想要一种更好的方法，但是np.nanmean不允许权重。 np.average不允许使用选项来控制NaN的处理。

Answer 1

没有比我的建议更明确的答案，我建议使用下面的功能还不错：

import pandas as pd
import numpy as np

def weighted_means_by_column_ignoring_NaNs(x, cols, w="weights"):
    """ This takes a DataFrame and averages each data column (cols),
        weighting observations by column w, but ignoring individual NaN
        observations within each column.
    """
    return pd.Series([np.nan if x.dropna(subset=[c]).empty else \
                      np.average(x.dropna(subset=[c])[c], 
                      weights =x.dropna(subset=[c])[w] )  \
                      for c in cols], cols)

用法示例如下

df=pd.DataFrame({'category':['a','a','b','b'],
 'var1':np.random.randint(0,100,4),
 'var2':np.random.randint(0,100,4),
 'weights':np.random.randint(0,10,4)})
df.loc[1,'var1']=np.nan
df


      category  var1  var2  weights
0        a      74.0    99        9
1        a       NaN     8        4
2        b      13.0    86        2
3        b      49.0    38        7

df.groupby('category').apply(weighted_means_by_column_ignoring_NaNs), 
        ['var1', 'var2'])


          var1       var2
category                 
a         74.0  57.846154
b         23.0   8.000000

Answer 2

如何将Nan值设置为零并创建一个新列var * weight。然后，您可以使用groupby来获得结果。

Answer 3

您可以在调用melt和调用dropna之前用apply和unstack预处理数据帧

wa=lambda x: np.average(x.value, weights=x.weights)
df_avg = (df.melt(['category', 'weights']).dropna().groupby(['category', 'variable'])
                                                   .apply(wa).unstack())

Out[40]:
variable  var1       var2
category
a         74.0  71.000000
b         41.0  48.666667

注意：您所需的输出与示例不匹配。 (a, 'var2')的值为(99 * 9 + 8 * 4) / (9 + 4) = 71

使用groupby的多列加权平均值，逐列丢弃NaNs

3 个答案: