PySpark Boolean Pivot

Question

I have some data mimicking the following structure:

rdd = sc.parallelize(
    [
        (0,1), 
        (0,5), 
        (0,3), 
        (1,2), 
        (1,3), 
        (2,6)
    ]
)

df_data = sqlContext.createDataFrame(rdd, ["group","value"])

df_data.show()

+-----+-----+
|group|value|
+-----+-----+
|    0|    1|
|    0|    5|
|    0|    3|
|    1|    2|
|    1|    3|
|    2|    6|
+-----+-----+

What I would like to do is to pivot this data by group to show the presence of the 'value' values as follows:

+-----+-------+-------+-------+-------+-------+
|group|value_1|value_2|value_3|value_5|value_6|
+-----+-------+-------+-------+-------+-------+
|    0|   true|  false|   true|   true|  false|
|    1|  false|   true|   true|  false|  false|
|    2|  false|  false|  false|  false|   true|
+-----+-------+-------+-------+-------+-------+

Is there any way I could achieve this with PySpark?

I have tried a combination of groupby/pivot/agg without any success.

Answer 1

@Psidom's answer will only work on Spark version 2.3 and higher as the pyspark.sql.DataFrameNaFunctions did not support bool in prior versions.

This is what I get when I run that code in Spark 2.1:

import pyspark.sql.functions as F

(df_data.withColumn('value', F.concat(F.lit('value_'), df_data.value))
        .groupBy('group').pivot('value').agg(F.count('*').isNotNull())
        .na.fill(False).show())
#+-----+-------+-------+-------+-------+-------+
#|group|value_1|value_2|value_3|value_5|value_6|
#+-----+-------+-------+-------+-------+-------+
#|    0|   true|   null|   true|   true|   null|
#|    1|   null|   true|   true|   null|   null|
#|    2|   null|   null|   null|   null|   true|
#+-----+-------+-------+-------+-------+-------+

---

Here is an alternative solution that should work for Spark 2.2 and lower:

# first pivot and fill nulls with 0
df = df_data.groupBy('group').pivot('value').count().na.fill(0)
df.show()
#+-----+---+---+---+---+---+
#|group|  1|  2|  3|  5|  6|
#+-----+---+---+---+---+---+
#|    0|  1|  0|  1|  1|  0|
#|    1|  0|  1|  1|  0|  0|
#|    2|  0|  0|  0|  0|  1|
#+-----+---+---+---+---+---+

Now use select to rename the columns and cast the values from int to bool:

df.select(
    *[F.col(c) if c == 'group' else F.col(c).cast('boolean').alias('value_'+c) 
      for c in df.columns]
).show()
+-----+-------+-------+-------+-------+-------+
|group|value_1|value_2|value_3|value_5|value_6|
+-----+-------+-------+-------+-------+-------+
|    0|   true|  false|   true|   true|  false|
|    1|  false|   true|   true|  false|  false|
|    2|  false|  false|  false|  false|   true|
+-----+-------+-------+-------+-------+-------+

Answer 2

Here is one way:

import pyspark.sql.functions as F

(df_data.withColumn('value', F.concat(F.lit('value_'), df_data.value))
        .groupBy('group').pivot('value').agg(F.count('*').isNotNull())
        .na.fill(False).show())

+-----+-------+-------+-------+-------+-------+
|group|value_1|value_2|value_3|value_5|value_6|
+-----+-------+-------+-------+-------+-------+
|    0|   true|  false|   true|   true|  false|
|    1|  false|   true|   true|  false|  false|
|    2|  false|  false|  false|  false|   true|
+-----+-------+-------+-------+-------+-------+