[SPARK-21190] SPIP: Vectorized UDFs in Python

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

[SPARK-21190] SPIP: Vectorized UDFs in Python

rxin
Welcome to the first real SPIP.

SPIP: Vectorized UDFs for Python

 

https://issues.apache.org/jira/browse/SPARK-21190

 

Background and Motivation:

 

Python is one of the most popular programming languages among Spark users. Spark currently exposes a row-at-a-time interface for defining and executing user-defined functions (UDFs). This introduces high overhead in serialization and deserialization, and also makes it difficult to leverage Python libraries (e.g. numpy, Pandas) that are written in native code.

 

This proposal advocates introducing new APIs to support vectorized UDFs in Python, in which a block of data is transferred over to Python in some columnar format for execution.

 

 

Target Personas:

Data scientists, data engineers, library developers.

 

Goals:

  • Support vectorized UDFs that apply on chunks of the data frame

  • Low system overhead: Substantially reduce serialization and deserialization overhead when compared with row-at-a-time interface

  • UDF performance: Enable users to leverage native libraries in Python (e.g. numpy, Pandas) for data manipulation in these UDFs

 

 

Non-Goals:

 

The following are explicitly out of scope for the current SPIP, and should be done in future SPIPs. Nonetheless, it would be good to consider these future use cases during API design, so we can achieve some consistency when rolling out new APIs.

 

- Define block oriented UDFs in other languages (that are not Python).

- Define aggregate UDFs

- Tight integration with machine learning frameworks

 

 

Proposed API Changes:

 

The following sketches some possibilities. I haven’t spent a lot of time thinking about the API (wrote it down in 5 mins) and I am not attached to this design at all. The main purpose of the SPIP is to get feedback on use cases and see how they can impact API design.

 

Some things to consider are:

 

1. Python is dynamically typed, whereas DataFrames/SQL requires static, analysis time typing. This means users would need to specify the return type of their UDFs.

 

2. Ratio of input rows to output rows. We propose initially we require number of output rows to be the same as the number of input rows. In the future, we can consider relaxing this constraint with support for vectorized aggregate UDFs.

 

3. On the Python side, what do we expose? The common libraries are numpy (more for numeric / machine learning), and Pandas (more for structured data analysis).

 

 

Proposed API sketch (using examples):

 

Use case 1. A function that defines all the columns of a DataFrame (similar to a “map” function):

 

@spark_udf(some way to describe the return schema)

def my_func_on_entire_df(input):

 """ Some user-defined function.

 

 :param input: A Pandas DataFrame with two columns, a and b.

 :return: :class: A Pandas data frame.

 """

 input[c] = input[a] + input[b]

 Input[d] = input[a] - input[b]

 return input

 

spark.range(1000).selectExpr("id a", "id / 2 b")

 .mapBatches(my_func_on_entire_df)

 

 

Use case 2. A function that defines only one column (similar to existing UDFs):

 

@spark_udf(some way to describe the return schema)

def my_func_that_returns_one_column(input):

 """ Some user-defined function.

 

 :param input: A Pandas DataFrame with two columns, a and b.

 :return: :class: A numpy array

 """

 return input[a] + input[b]

 

my_func = udf(my_func_that_returns_one_column)

 

df = spark.range(1000).selectExpr("id a", "id / 2 b")

df.withColumn("c", my_func(df.a, df.b))

 

 

Optional Design Sketch:

 

I’m more concerned about getting proper feedback for API design. The implementation should be pretty straightforward and is not a huge concern at this point. We can leverage the same implementation for faster toPandas (using Arrow).

 

 

Optional Rejected Designs:

 

See above.




Loading...