PySpark reduceByKey aggregation after collect_list

Question

I want to take following example to do my aggregation according to 'states' collected by collect_list.

example code:

states = sc.parallelize(["TX","TX","CA","TX","CA"])
states.map(lambda x:(x,1)).reduceByKey(operator.add).collect()
#printed output: [('TX', 3), ('CA', 2)]

my code:

from pyspark import SparkContext,SparkConf
from pyspark.sql.session import SparkSession
from pyspark.sql.functions import collect_list
import operator
conf = SparkConf().setMaster("local")
conf = conf.setAppName("test")
sc = SparkContext.getOrCreate(conf=conf)
spark = SparkSession(sc)
rdd = sc.parallelize([('20170901',['TX','TX','CA','TX']), ('20170902', ['TX','CA','CA']), ('20170902',['TX']) ])
df = spark.createDataFrame(rdd, ["datatime", "actionlist"])
df = df.groupBy("datatime").agg(collect_list("actionlist").alias("actionlist"))

rdd = df.select("actionlist").rdd.map(lambda x:(x,1))#.reduceByKey(operator.add)
print (rdd.take(2))
#printed output: [(Row(actionlist=[['TX', 'CA', 'CA'], ['TX']]), 1 (Row(actionlist=[['TX', 'TX', 'CA', 'TX']]), 1)]
#for next step, it should look like:
#[(Row(actionlist=[('TX',1), ('CA',1), ('CA',1), ('TX',1)]), (Row(actionlist=[('TX',1), ('TX',1), ('CA',1), ('TX',1)])]

what I want is something like:

20170901,[('TX', 3), ('CA', 1 )]
20170902,[('TX', 2), ('CA', 2 )]

I think first step is to flatten collect_list result, I've tried: udf(lambda x: list(chain.from_iterable(x)), StringType()) udf(lambda items: list(chain.from_iterable(itertools.repeat(x,1) if isinstance(x,str) else x for x in items))) udf(lambda l: [item for sublist in l for item in sublist])

but no luck yet, next step is to makeup KV pairs and do reduce, I stuck here for a while, can any spark expert help on the logic? appreciate you help!

Answer 1

You can use reduce and counter in udf to achieve it. I tried my way, hope this helps.

>>> from functools import reduce
>>> from collections import Counter
>>> from pyspark.sql.types import *
>>> from pyspark.sql import functions as F
>>> rdd = sc.parallelize([('20170901',['TX','TX','CA','TX']), ('20170902', ['TX','CA','CA']), ('20170902',['TX']) ])
>>> df = spark.createDataFrame(rdd, ["datatime", "actionlist"])
>>> df = df.groupBy("datatime").agg(F.collect_list("actionlist").alias("actionlist"))
>>> def someudf(row):
        value = reduce(lambda x,y:x+y,row)
        return Counter(value).most_common()

>>> schema = ArrayType(StructType([
    StructField("char", StringType(), False),
    StructField("count", IntegerType(), False)]))

>>> udf1 = F.udf(someudf,schema)
>>> df.select('datatime',udf1(df.actionlist)).show(2,False)
+--------+-------------------+
|datatime|someudf(actionlist)|
+--------+-------------------+
|20170902|[[TX,2], [CA,2]]   |
|20170901|[[TX,3], [CA,1]]   |
+--------+-------------------+

Answer 2

You can simply do ,it by using combineByKey():

from collections import Counter
count = rdd.combineByKey(lambda v: Counter(v),
                                 lambda c,v: c + Counter(v),
                                 lambda c1,c2: c1 + c2)
print count #[('20170901', Counter({'TX': 3, 'CA': 1})), ('20170902', Counter({'CA': 2, 'TX': 2}))]