PySpark - 将单个整数列表与列表列进行比较

Question

I'm trying to check which entries in a spark dataframe (column with lists) contain the largest quantity of values from a given list.

The best approach I've came up with is iterating over a dataframe with rdd.foreach() and comparing a given list to every entry using python's set1.intersection(set2) .

My question is does spark have any built-in functionality for this so iterating with .foreach could be avoided?

Thanks for any help!

PS my dataframe looks like this:

+-------------+---------------------+                                           
|   cardnumber|collect_list(article)|
+-------------+---------------------+
|2310000000855| [12480, 49627, 80...|
|2310000008455| [35531, 22564, 15...|
|2310000011462| [117112, 156087, ...|
+-------------+---------------------+

And I'm trying to find entries with the most intersections in the second column with a given list of articles, eg [151574, 87239, 117908, 162475, 48599]

Answer 1

You can try the same set operation in dataframe instead of using rdd.foreach:

from pyspark.sql.functions import udf, li, col
my_udf=udf(lambda A,B: list(set(A).intersection(set(B))))
df=df.withColumn('intersect_value', my_udf('A', 'B'))

You can use the len function to get the size of intersect list in the UDF itself and perform the operation you want from this dataframe.

Answer 2

The only alternative here is udf , but it won't be much of a difference.

from pyspark.sql.functions import udf, li, col

def intersect(xs):
    xs = set(xs)
    @udf("array<long>")
    def _(ys):
        return list(xs.intersection(ys))
    return _

It can be applied as:

a_list = [1, 4, 6]

df = spark.createDataFrame([
    (1, [3, 4, 8]), (2, [7, 2, 6])
], ("id", "articles"))

df.withColumn("intersect", intersect(a_list)("articles")).show()

# +---+---------+---------+
# | id| articles|intersect|
# +---+---------+---------+
# |  1|[3, 4, 8]|      [4]|
# |  2|[7, 2, 6]|      [6]|
# +---+---------+---------+

Based on the names, it looks like you use collect_list so your data looks probably like this:

df_long = spark.createDataFrame([
    (1, 3),(1, 4), (1, 8), (2, 7), (2, 7), (2, 6)
], ("id", "articles"))

In that case problem is simpler. Join

lookup = spark.createDataFrame(a_list, "long").toDF("articles")

joined = lookup.join(df_long, ["articles"])

and aggregate the result:

joined.groupBy("id").count().show()
# +---+-----+                                                                     
# | id|count|
# +---+-----+
# |  1|    1|
# |  2|    1|
# +---+-----+


joined.groupBy("id").agg(collect_list("articles")).show()
# +---+----------------------+                                                    
# | id|collect_list(articles)|
# +---+----------------------+
# |  1|                   [4]|
# |  2|                   [6]|
# +---+----------------------+