基于 PySpark 中另一个数据帧的列值创建指标数组

我有两个数据框：df1

+---+-----------------+
|id1|           items1|
+---+-----------------+
|  0|     [B, C, D, E]|
|  1|        [E, A, C]|
|  2|     [F, A, E, B]|
|  3|        [E, G, A]|
|  4|  [A, C, E, B, D]|
+---+-----------------+ 

and df2:

+---+-----------------+
|id2|           items2|
+---+-----------------+
|001|           [A, C]|
|002|              [D]|
|003|        [E, A, B]|
|004|        [B, D, C]|
|005|           [F, B]|
|006|           [G, E]|
+---+-----------------+ 

我想创建一个指示向量（在新列中result_array in df1）基于中的值items2。向量的长度应与中的行数相同df2（在本例中它应该有 6 个元素）。如果行位于items1包含相应行中的所有元素items2，否则值为 0.0。结果应如下所示：

+---+-----------------+-------------------------+
|id1|           items1|             result_array|
+---+-----------------+-------------------------+
|  0|     [B, C, D, E]|[0.0,1.0,0.0,1.0,0.0,0.0]|
|  1|        [E, A, C]|[1.0,0.0,0.0,0.0,0.0,0.0]|
|  2|     [F, A, E, B]|[0.0,0.0,1.0,0.0,1.0,0.0]|
|  3|        [E, G, A]|[0.0,0.0,0.0,0.0,0.0,1.0]|
|  4|  [A, C, E, B, D]|[1.0,1.0,1.0,1.0,0.0,0.0]|
+---+-----------------+-------------------------+

例如，在第 0 行中，第二个值为 1.0，因为 [D] 是 [B, C, D, E] 的子集，第四个值为 1.0，因为 [B, D, C] 是 [B, D, C] 的子集， C、D、E]。中的所有其他项目组df2不是[B,C,D,E]的子集，因此它们的指标值为0.0。

我尝试创建所有项目组的列表items2使用collect()，然后应用udf，但我的数据太大（超过1000万行）。

你可以这样继续，

import pyspark.sql.functions as F
from pyspark.sql.types import *

df1 = sql.createDataFrame([
     (0,['B', 'C', 'D', 'E']),
     (1,['E', 'A', 'C']),
     (2,['F', 'A', 'E', 'B']),
     (3,['E', 'G', 'A']),
     (4,['A', 'C', 'E', 'B', 'D'])],
   ['id1','items1'])

df2 = sql.createDataFrame([
     (001,['A', 'C']),
     (002,['D']),
     (003,['E', 'A', 'B']),
     (004,['B', 'D', 'C']),
     (005,['F', 'B']),
     (006,['G', 'E'])],
    ['id2','items2'])

它为您提供了数据框，

+---+---------------+
|id1|         items1|
+---+---------------+
|  0|   [B, C, D, E]|
|  1|      [E, A, C]|
|  2|   [F, A, E, B]|
|  3|      [E, G, A]|
|  4|[A, C, E, B, D]|
+---+---------------+

+---+---------+
|id2|   items2|
+---+---------+
|  1|   [A, C]|
|  2|      [D]|
|  3|[E, A, B]|
|  4|[B, D, C]|
|  5|   [F, B]|
|  6|   [G, E]|
+---+---------+

Now, crossJoin这两个数据框，为您提供笛卡尔积df1 with df2. Then, groupby on 'items1'并应用一个udf得到'result_array'.

get_array_udf = F.udf(lambda x,y:[1.0 if set(z) < set(x) else 0.0 for z in y], ArrayType(FloatType()))

df = df1.crossJoin(df2)\
        .groupby(['id1', 'items1']).agg(F.collect_list('items2').alias('items2'))\
        .withColumn('result_array', get_array_udf('items1', 'items2')).drop('items2')

df.show()

这会给你输出，

+---+---------------+------------------------------+                            
|id1|items1         |result_array                  |
+---+---------------+------------------------------+
|1  |[E, A, C]      |[1.0, 0.0, 0.0, 0.0, 0.0, 0.0]|
|0  |[B, C, D, E]   |[0.0, 1.0, 0.0, 1.0, 0.0, 0.0]|
|4  |[A, C, E, B, D]|[1.0, 1.0, 1.0, 1.0, 0.0, 0.0]|
|3  |[E, G, A]      |[0.0, 0.0, 0.0, 0.0, 0.0, 1.0]|
|2  |[F, A, E, B]   |[0.0, 0.0, 1.0, 0.0, 1.0, 0.0]|
+---+---------------+------------------------------+