Remove an element from a Python list of lists in PySpark DataFrame

Question

I am trying to remove an element from a Python list of lists:

+---------------+
|        sources|
+---------------+
|           [62]|
|        [7, 32]|
|           [62]|
|   [18, 36, 62]|
|[7, 31, 36, 62]|
|    [7, 32, 62]|

I want to be able to remove an element, rm , from each of the lists in the list above. I wrote a function that can do that for a list of lists:

def asdf(df, rm):
    temp = df
    for n in range(len(df)):
        temp[n] = [x for x in df[n] if x != rm]
    return(temp)

which does remove rm = 1 :

a = [[1,2,3],[1,2,3,4],[1,2,3,4,5]]
In:  asdf(a,1)
Out: [[2, 3], [2, 3, 4], [2, 3, 4, 5]]

But I can't get it to work for a DataFrame:

asdfUDF = udf(asdf, ArrayType(IntegerType()))

In: df.withColumn("src_ex", asdfUDF("sources", 32))

Out: Py4JError: An error occurred while calling z:org.apache.spark.sql.functions.col. Trace:
py4j.Py4JException: Method col([class java.lang.Integer]) does not exist

Desired behavior:

In: df.withColumn("src_ex", asdfUDF("sources", 32))
Out: 

+---------------+
|         src_ex|
+---------------+
|           [62]|
|            [7]|
|           [62]|
|   [18, 36, 62]|
|[7, 31, 36, 62]|
|        [7, 62]|

(except have the new column above appended to a PySpark DataFrame, df )

Any suggestions or ideas?

Answer 1

Spark >= 2.4

You can use array_remove :

from pyspark.sql.functions import array_remove

df.withColumn("src_ex", array_remove("sources", 32)).show()

+---------------+---------------+
|        sources|         src_ex|
+---------------+---------------+
|           [62]|           [62]|
|        [7, 32]|            [7]|
|           [62]|           [62]|
|   [18, 36, 62]|   [18, 36, 62]|
|[7, 31, 36, 62]|[7, 31, 36, 62]|
|    [7, 32, 62]|        [7, 62]|
+---------------+---------------+

or filter :

from pyspark.sql.functions import expr

df.withColumn("src_ex", expr("filter(sources, x -> not(x <=> 32))")).show()

+---------------+---------------+
|        sources|         src_ex|
+---------------+---------------+
|           [62]|           [62]|
|        [7, 32]|            [7]|
|           [62]|           [62]|
|   [18, 36, 62]|   [18, 36, 62]|
|[7, 31, 36, 62]|[7, 31, 36, 62]|
|    [7, 32, 62]|        [7, 62]|
+---------------+---------------+

Spark < 2.4

A number of things:

• DataFrame is not a list of lists . In practice it is not even a plain Python object, it has no len and it is not Iterable .
• Column you have looks like plain array type.
• You cannot reference DataFrame (or any other distributed data structure inside UDF).
• Every argument passed directly to UDF call has to be a str (column name) or Column object. To pass literal use lit function.

The only thing that remains is just a list comprehension:

from pyspark.sql.functions import lit, udf

def drop_from_array_(arr, item):
    return [x for x in arr if x != item]

drop_from_array = udf(drop_from_array_, ArrayType(IntegerType()))

Example usage:

df = sc.parallelize([
    [62], [7, 32], [62], [18, 36, 62], [7, 31, 36, 62], [7, 32, 62]
]).map(lambda x: (x, )).toDF(["sources"])

df.withColumn("src_ex", drop_from_array("sources", lit(32)))

The result:

+---------------+---------------+
|        sources|         src_ex|
+---------------+---------------+
|           [62]|           [62]|
|        [7, 32]|            [7]|
|           [62]|           [62]|
|   [18, 36, 62]|   [18, 36, 62]|
|[7, 31, 36, 62]|[7, 31, 36, 62]|
|    [7, 32, 62]|        [7, 62]|
+---------------+---------------+