CODEWITHSUNDEEP
apache-sparkdataframepysparkuser-defined-functions
Michail N
Michail N

Reputation: 3845

Spark dataframe operation on list returns [Ljava.lang.Object;@]

I have a pyspark dataframe where I have grouped data to list with collect_list.

from pyspark.sql.functions import udf, collect_list
from itertools import combinations, chain

#Create Dataframe
df = spark.createDataFrame( [(1,'a'), (1,'b'), (2,'c')] , ["id", "colA"])   

df.show()
>>>
+---+----+
| id|colA|
+---+----+
|  1|   a|
|  1|   b|
|  2|   c|
+---+----+

#Group by and collect to list
df = df.groupBy(df.id).agg(collect_list("colA").alias("colAlist"))

df.show()
>>>
+---+--------+
| id|colAList|
+---+--------+
|  1|  [a, b]|
|  2|     [c]|
+---+--------+

Then I use a function to find all combinations of the list elements to a new list 

allsubsets = lambda l: list(chain(*[combinations(l , n) for n in range(1,len(l)+1)]))
df = df.withColumn('colAsubsets',udf(allsubsets)(df['colAList']))

so I would excpect something like 

+---+--------------------+
| id| colAsubsets        |
+---+--------------------+
|  1|  [[a], [b], [a,b]] |
|  2|  [[b]]             |
+---+--------------------+

but I get:

df.show()
>>>
 +---+--------+-----------------------------------------------------------------------------------------+
|id |colAList|colAsubsets                                                                              |
+---+--------+-----------------------------------------------------------------------------------------+
|1  |[a, b]  |[[Ljava.lang.Object;@75e2d657, [Ljava.lang.Object;@7f662637, [Ljava.lang.Object;@b572639]|
|2  |[c]     |[[Ljava.lang.Object;@26f67148]                                                           |
+---+--------+-----------------------------------------------------------------------------------------+

Any ideas what to do? And then maybe how to flatten the list to different rows?

Upvotes: 3

Views: 1692

Answers (2)

Ramesh Maharjan
Ramesh Maharjan

Reputation: 41957

All you need to do is to extract the elements from objects created by chain and combinations in a flattened way

so changing 

allsubsets = lambda l: list(chain(*[combinations(l , n) for n in range(1,len(l)+1)]))

to the following 

allsubsets = lambda l: [[z for z in y] for y in chain(*[combinations(l , n) for n in range(1,len(l)+1)])]

should give you 

+---+---------+------------------+
|id |colA_list|colAsubsets       |
+---+---------+------------------+
|1  |[a, b]   |[[a], [b], [a, b]]|
|2  |[c]      |[[c]]             |
+---+---------+------------------+

I hope the answer is helpful

Upvotes: 3

Pierre Gourseaud
Pierre Gourseaud

Reputation: 2477

Improving on @RameshMaharjan answer, in order to flatten the list to different rows:

You have to use explode on an array. You must before specify the type of your udf so it doesn't return a StringType.

from pyspark.sql.functions import explode
from pyspark.sql.types import ArrayType, StringType

allsubsets = lambda l: [[z for z in y] for y in chain(*[combinations(l , n) for n in range(1,len(l)+1)])]
df = df.withColumn('colAsubsets', udf(allsubsets, ArrayType(ArrayType(StringType())))(df['colAList']))
df = df.select('id', explode('colAsubsets'))

Result :

+---+------+
| id|   col|
+---+------+
|  1|   [a]|
|  1|   [b]|
|  1|[a, b]|
|  2|   [c]|
+---+------+

Upvotes: 3

Related Questions