如何获取具有Pyspark Dataframe的另一列中给出的多个列的值的列表列?

时间:2019-07-18 03:35:33

标签: python pyspark apache-spark-sql pyspark-sql

有没有办法创建一个新列,例如下面在Pyspark中显示的Dataframe?

我一直在尝试列表理解:

import pyspark.functions as F

df.withColumn('result', [F.col(colname) for colname in F.col('colList')])

但是不起作用。

预期结果是:

+----+----+----+----+----+---------------+------+
|col1|col2|col3|col4|col5|        colList|result|
+----+----+----+----+----+---------------+------+
|   1|   2|   0|   3|   4|['col1','col2']| [1,2]|
|   1|   2|   0|   3|   4|['col2','col3']| [2,0]|
|   1|   2|   0|   3|   4|['col1','col3']| [1,0]|
|   1|   2|   0|   3|   4|['col3','col4']| [0,3]|
|   1|   2|   0|   3|   4|['col2','col5']| [2,4]|
|   1|   2|   0|   3|   4|['col4','col5']| [3,4]|
+----+----+----+----+----+---------------+------+

1 个答案:

答案 0 :(得分:0)

# Loading requisite functions and creating the DataFrame
from pyspark.sql.functions import create_map, lit, col, struct
from itertools import chain

myValues = [(1,2,0,3,4,['col1','col2']),(1,2,0,3,4,['col2','col3']),
            (1,2,0,3,4,['col1','col3']),(1,2,0,3,4,['col3','col4']),
            (1,2,0,3,4,['col2','col5']),(1,2,0,3,4,['col4','col5'])]
df = sqlContext.createDataFrame(myValues,['col1','col2','col3','col4','col5','colList'])
df.show()
+----+----+----+----+----+------------+
|col1|col2|col3|col4|col5|     colList|
+----+----+----+----+----+------------+
|   1|   2|   0|   3|   4|[col1, col2]|
|   1|   2|   0|   3|   4|[col2, col3]|
|   1|   2|   0|   3|   4|[col1, col3]|
|   1|   2|   0|   3|   4|[col3, col4]|
|   1|   2|   0|   3|   4|[col2, col5]|
|   1|   2|   0|   3|   4|[col4, col5]|
+----+----+----+----+----+------------+

下一步,我们为colList数组中的各个列创建列。

df = df.withColumn('first_col',col('colList')[0])
df = df.withColumn('second_col',col('colList')[1])
df.show()
+----+----+----+----+----+------------+---------+----------+
|col1|col2|col3|col4|col5|     colList|first_col|second_col|
+----+----+----+----+----+------------+---------+----------+
|   1|   2|   0|   3|   4|[col1, col2]|     col1|      col2|
|   1|   2|   0|   3|   4|[col2, col3]|     col2|      col3|
|   1|   2|   0|   3|   4|[col1, col3]|     col1|      col3|
|   1|   2|   0|   3|   4|[col3, col4]|     col3|      col4|
|   1|   2|   0|   3|   4|[col2, col5]|     col2|      col5|
|   1|   2|   0|   3|   4|[col4, col5]|     col4|      col5|
+----+----+----+----+----+------------+---------+----------+

具有整数值的列的列表-

concerned_columns = [x for x in df.columns if x not in {'colList','first_col','second_col'}]
print(concerned_columns)
    ['col1', 'col2', 'col3', 'col4', 'col5']

现在,最重要的部分是,我们使用create_map函数创建了列名称及其相应值之间的映射,此函数已在Spark 2. +及更高版本中提供。

# Maping - (column name, column values)
col_name_value_mapping = create_map(*chain.from_iterable(
    (lit(c), col(c)) for c in concerned_columns
))

最后,应用此映射来获取存储在列 first_col second_col 中的列的值,并使用struct将它们放入数组中。

df = df.withColumn('result', struct(col_name_value_mapping[col('first_col')],col_name_value_mapping[col('second_col')]))
df = df.drop('first_col','second_col')
df.show()
+----+----+----+----+----+------------+------+
|col1|col2|col3|col4|col5|     colList|result|
+----+----+----+----+----+------------+------+
|   1|   2|   0|   3|   4|[col1, col2]| [1,2]|
|   1|   2|   0|   3|   4|[col2, col3]| [2,0]|
|   1|   2|   0|   3|   4|[col1, col3]| [1,0]|
|   1|   2|   0|   3|   4|[col3, col4]| [0,3]|
|   1|   2|   0|   3|   4|[col2, col5]| [2,4]|
|   1|   2|   0|   3|   4|[col4, col5]| [3,4]|
+----+----+----+----+----+------------+------+