Append to PySpark array column

Question

I want to check if the column values are within some boundaries. If they are not I will append some value to the array column "F". This is the code I have so far:

df = spark.createDataFrame(
    [
        (1, 56), 
        (2, 32),
        (3, 99)
    ],
    ['id', 'some_nr'] 
)

df = df.withColumn( "F", F.lit( None ).cast( types.ArrayType( types.ShortType( ) ) ) )

def boundary_check( val ):
  if (val > 60) | (val < 50):
    return 1

udf  = F.udf( lambda x: boundary_check( x ) ) 

df =  df.withColumn("F", udf(F.col("some_nr")))
display(df)

However, I don't know how to append to the array. Currently, if I perform another boundary check on df it will simply overwrite the previous values in "F"...

Napoleon Borntoparty · Accepted Answer · 2020-01-21T13:55:40.833

Have a look at the array_union function under pyspark.sql.functions here: https://spark.apache.org/docs/latest/api/python/pyspark.sql.html?highlight=join#pyspark.sql.functions.array_union

That way you avoid using udf, which takes away any benefits of Spark parallelisation. The code would look something like:

from pyspark.context import SparkContext
from pyspark.sql import SparkSession
from pyspark.conf import SparkConf
from pyspark.sql import Row
import pyspark.sql.functions as f


conf = SparkConf()
sc = SparkContext(conf=conf)
spark = SparkSession(sc)

df = spark.createDataFrame([Row(c1=["b", "a", "c"], c2="a", c3=10),
                            Row(c1=["b", "a", "c"], c2="d", c3=20)])
df.show()
+---------+---+---+
|       c1| c2| c3|
+---------+---+---+
|[b, a, c]|  a| 10|
|[b, a, c]|  d| 20|
+---------+---+---+

df.withColumn(
    "output_column", 
    f.when(f.col("c3") > 10, 
           f.array_union(df.c1, f.array(f.lit("1"))))
     .otherwise(f.col("c1"))
).show()
+---------+---+---+-------------+
|       c1| c2| c3|output_column|
+---------+---+---+-------------+
|[b, a, c]|  a| 10|    [b, a, c]|
|[b, a, c]|  d| 20| [b, a, c, 1]|
+---------+---+---+-------------+

As as side note, this works as a logical union, therefore if you want to append a value, you need to make sure this value is unique so that it always gets added. Otherwise, have a look at other array functions here: https://spark.apache.org/docs/latest/api/python/pyspark.sql.html?highlight=join#pyspark.sql.functions.array

NB: Your spark needs to be version >2.4 for most of the array functions.

EDIT (on request in comments):

The withColumn method only allows you to work on one column at a time, so you need to use a new withColumn, ideally with predefining your logical statement ahead for both withColumn queries.

logical_gate = (f.col("c3") > 10)

(
    df.withColumn(
        "output_column", 
        f.when(logical_gate, 
               f.array_union(df.c1, f.array(f.lit("1"))))
         .otherwise(f.col("c1")))
      .withColumn(
        "c3",
        f.when(logical_gate,
               f.lit(None))
         .otherwise(f.col("c3")))
      .show()
)
+---------+---+----+-------------+
|       c1| c2|  c3|output_column|
+---------+---+----+-------------+
|[b, a, c]|  a|  10|    [b, a, c]|
|[b, a, c]|  d|null| [b, a, c, 1]|
+---------+---+----+-------------+

Works great, thanks! Is there a way to have two actions on one when clause? Say when c3 > 10 append 1 to array c1 AND replace c3 with None — Lossa, Jan 21 '20 at 13:42

score 0 · Answer 2 · answered May 12 '23 at 13:47

Since Spark 3.4+ you can use array_append:

from pyspark.sql import functions as F

df = spark.createDataFrame([(10, ['a', 'b', 'c']), (20, ['a', 'b', 'c'])], ['c1', 'c2'])
df.show()
# +---+---------+
# | c1|       c2|
# +---+---------+
# | 10|[a, b, c]|
# | 20|[a, b, c]|
# +---+---------+

df = df.withColumn('c2', F.when(F.col('c1') > 15, F.array_append('c2', 'd')).otherwise(F.col('c2')))
df.show()
# +---+------------+
# | c1|          c2|
# +---+------------+
# | 10|   [a, b, c]|
# | 20|[a, b, c, d]|
# +---+------------+

Append to PySpark array column

2 Answers2