在多列Pyspark上进行Groupby操作

Question

我已经应用了一个groupby并计算了pyspark数据帧中两个特征的标准差

from pyspark.sql import functions as f


val1 = [('a',20,100),('a',100,100),('a',50,100),('b',0,100),('b',0,100),('c',0,0),('c',0,50),('c',0,100),('c',0,20)]
cols = ['group','val1','val2']
tf = spark.createDataFrame(val1, cols)
tf.show() 
tf.groupby('group').agg(f.stddev(['val1','val2']).alias('val1_std','val2_std'))

但它给了我以下错误

TypeError: _() takes 1 positional argument but 2 were given

如何在pyspark中执行？

Answer 1

问题是stddev函数作用于单个列而不是像您编写的代码中的多个列（因此关于1对2参数的错误消息）。获得所需内容的一种方法是分别计算每列的标准偏差：

# std dev for each col
expressions = [f.stddev(col).alias('%s_std'%(col)) for col in ['val1','val2']]
# Now run it
tf.groupby('group').agg(*expressions).show()

#+-----+------------------+------------------+
#|group|          val1_std|          val2_std|
#+-----+------------------+------------------+
#|    c|               0.0|43.493294502332965|
#|    b|               0.0|               0.0|
#|    a|40.414518843273804|               0.0|
#+-----+------------------+------------------+