我正在尝试计算pyspark中的加权均值,但没有取得很大进展

# Example data
df = sc.parallelize([
    ("a", 7, 1), ("a", 5, 2), ("a", 4, 3),
    ("b", 2, 2), ("b", 5, 4), ("c", 1, -1)
]).toDF(["k", "v1", "v2"])
df.show()

import numpy as np
def weighted_mean(workclass, final_weight):
    return np.average(workclass, weights=final_weight)

weighted_mean_udaf = pyspark.sql.functions.udf(weighted_mean,
    pyspark.sql.types.IntegerType())

但是当我尝试执行此代码时

df.groupby('k').agg(weighted_mean_udaf(df.v1,df.v2)).show()

我收到错误

u"expression 'pythonUDF' is neither present in the group by, nor is it an aggregate function. Add to group by or wrap in first() (or first_value) if you don't care which value you get

我的问题是,我可以指定一个自定义函数(采用多个参数)作为agg的参数吗？如果不是,是否有其他选择可以在按键分组后执行加权平均等操作？

解决方法:

用户定义的聚合功能(UDAF在pyspark.sql.GroupedData上有效,但在pyspark中不受支持)不是用户定义的函数(UDF,在pyspark.sql.DataFrame上有效).

因为在pyspark中您无法创建自己的UDAF,并且提供的UDAF无法解决您的问题,因此您可能需要返回RDD世界：

from numpy import sum

def weighted_mean(vals):
    vals = list(vals)  # save the values from the iterator
    sum_of_weights = sum(tup[1] for tup in vals)
    return sum(1. * tup[0] * tup[1] / sum_of_weights for tup in vals)

df.map(
    lambda x: (x[0], tuple(x[1:]))  # reshape to (key, val) so grouping could work
).groupByKey().mapValues(
    weighted_mean
).collect()