Handling Aggregations in Feature Function

[NaeemS]

Hi,

Is it possible to cater aggregation using Feature Functions somehow. As we know that the logic defined in feature function is applied on a single row when a join is being performed. But do we have any mechanism to handle to aggregations too somehow in our pipeline. We have methods like Pandas UDF ( with group by ) and UDAFs in Scala. Can we somehow leverage any of those for logging the group by and other conditions to automate our whole pipeline including the grouping and filtering conditions.

Thanks in Advance!

[Kaniz_Fatma]

Hi @NaeemS, 

Pandas UDFs in PySpark can be used for vectorizing scalar operations, which can significantly improv...¹. You can use them with APIs such as select and withColumn. The Python function should take a pandas Series as an input and return a pandas Series of the same s...¹. For example, you can define a Pandas UDF that computes the product of 2 columns¹. Also, you can use groupby().apply() with Pandas UDFs².

UDAFs in Scala allow you to perform custom calculations over groups of input data as part of Spark S...³. For example, you can define a UDAF that calculates the geometric mean of a group of values.

In both cases, you can log the group by and other conditions to automate your whole pipeline including the grouping and filtering conditions.

[NaeemS]

Hi @Kaniz_Fatma ,

Thanks for your reply. I'm familiar with both of these. But I was wondering if can include that part while logging our pipeline using feature stores to handle the grouping and filtering as well.