How can I define user-defined aggregate functions in PySpark?

Question

I want make an user defined aggregate function in pyspark. I found some documentation for Scala and would like to achieve something similar in Python.

To be more specific, assume I already have a function like this implemented:

def process_data(df: pyspark.sql.DataFrame) -> bytes:
  ...  # do something very complicated here

and now I would like to be able to do something like:

source_df.groupBy("Foo_ID").agg(UDAF(process_data))

Now the question is - what should I put in place of UDAF?

Answer 1

PySpark does not support UDAF directly, so we have to do aggregation manually.

Reference:

1. https://spark.apache.org/docs/latest/sql-ref-functions-udf-aggregate.html

2. How to write Pyspark UDAF on multiple columns?

3. Applying UDFs on GroupedData in PySpark (with functioning python example)

4. https://florianwilhelm.info/2017/10/efficient_udfs_with_pyspark/