Using NumPy to replace Pandas GroupBy-Apply pattern for performance.
If you use PySpark a lot you would know that the DataFrame API is great. However there are times when it is not sufficient because it does not cover every single piece of functionality we may want. This is where the Pandas UDF functionality comes in. The nice thing about the Pandas UDF functionality is that it uses Arrow for data transfer between Spark and Pandas which minimizes serialization-deserialization costs. I have a slight preference for Pandas Function API over Pandas UDF but now let’s get to the meat of the post which is about speeding up the Pandas GroupBy-Apply pattern by using NumPy instead.
Read in full here:
This thread was posted by one of our members via one of our news source trackers.