Improving Pandas and PySpark performance and interoperability with Apache Arrow
2 likes1,650 views
The presentation discusses improving interoperability between Pandas and PySpark using Apache Arrow, highlighting PySpark's capabilities and current limitations, particularly with user-defined functions (UDFs). It details the advantages of Apache Arrow, including faster data processing and efficient memory usage, and showcases performance improvements seen with vectorized UDFs. The talk concludes with future plans for further enhancement in PySpark and Pandas integration through Arrow.
![Columnar Data Format
persons = [{
name: ’Joe',
age: 18,
phones: [
‘555-111-1111’,
‘555-222-2222’
]
}, {
name: ’Jack',
age: 37,
phones: [ ‘555-333-3333’
]
}]](https://image.slidesharecdn.com/ytdhj6hrqzb9qkdbjaha-signature-3730a22e327b32f1c622ad8f856f39f96036318b91e17e82f9d13305b13fcd15-poli-180119221721/85/Improving-Pandas-and-PySpark-performance-and-interoperability-with-Apache-Arrow-20-320.jpg){kind=icon}
![Record Batch Construction
Schema
Dictionary
Batch
Record
Batch
Record
Batch
Record
Batch
name (offset)
name (data)
age (data)
phones (list offset)
phones (data)
data header (describes offsets into data)
name (bitmap)
age (bitmap)
phones (bitmap)
phones (offset)
{
name: ’Joe',
age: 18,
phones: [
‘555-111-1111’,
‘555-222-2222’
]
}
Each box (vector) is contiguous memory
The entire record batch is contiguous on wire](https://image.slidesharecdn.com/ytdhj6hrqzb9qkdbjaha-signature-3730a22e327b32f1c622ad8f856f39f96036318b91e17e82f9d13305b13fcd15-poli-180119221721/85/Improving-Pandas-and-PySpark-performance-and-interoperability-with-Apache-Arrow-21-320.jpg){kind=icon}
Improving Pandas and PySpark performance and interoperability with Apache Arrow
- 1. Improving Pandas and PySpark interoperability with Apache Arrow Li Jin PyData NYC November 2017
- 2. • The information presented here is offered for informational purposes only and should not be used for any other purpose (including, without limitation, the making of investment decisions). Examples provided herein are for illustrative purposes only and are not necessarily based on actual data. Nothing herein constitutes: an offer to sell or the solicitation of any offer to buy any security or other interest; tax advice; or investment advice. This presentation shall remain the property of Two Sigma Investments, LP (“Two Sigma”) and Two Sigma reserves the right to require the return of this presentation at any time. • Some of the images, logos or other material used herein may be protected by copyright and/or trademark. If so, such copyrights and/or trademarks are most likely owned by the entity that created the material and are used purely for identification and comment as fair use under international copyright and/or trademark laws. Use of such image, copyright or trademark does not imply any association with such organization (or endorsement of such organization) by Two Sigma, nor vice versa. • Copyright © 2017 TWO SIGMA INVESTMENTS, LP. All rights reserved IMPORTANT LEGAL INFORMATION
- 3. About Me 3 • Li Jin (@icexelloss) • Software Engineer @ Two Sigma Investments • Apache Arrow Committer • Analytics Tools Smith • Other Open Source Projects: • Flint: A Time Series Library on Spark • Cook: A Fair Scheduler on Mesos
- 4. • PySpark Overview • PySpark UDF: current state and limitation • Apache Arrow Overview • Improvement to PySpark UDF with Apache Arrow • Future Roadmap This Talk 4
- 6. • A tool for distributed data analysis • Apache project • JVM-based with Python interface (PySpark) • Functionality: • Relational: Join, group, aggregate … • Stats and ML: Spark MLlib • Streaming • … Apache Spark 6
- 7. • Bigger Data: • Pandas: 10G • Spark: 1000G • Better Parallelism: • Pandas: Single core • Spark: Hundreds of cores Why Spark 7
- 8. • Python interface for Spark • API front-end for built-in Spark functions • df.withColumn(‘v2’, df.v1 + 1) • Translated to Java code, running in JVM • Interface for native Python code (User-defined function) • df.withColumn(‘v2’, udf(lambda x: x+1, ‘double’)(df.v1)) • Running in Python runtime PySpark Overview 8
- 9. PySpark UDF: Current state and limitation 9
- 10. • PySpark’s interface to interact with other Python libraries • Types of UDFs: • Row UDF • Group UDF PySpark User Defined Function (UDF) 10
- 11. • Operates on row by row basis • Similar to `map` operator • Example: • String processing • Timestamp processing • Poor performance • 1-2 orders of magnitude slower comparing to alternatives (built-in Spark functions or vectorized operations) Row UDF: Current 11
- 12. • UDF that operates on multiple rows • Similar to `groupBy` followed by `map` operator • Example: • Monthly weighted mean • Not supported out of box • Poor performance Group UDF: Current 12
- 13. • (values – values.mean()) / values.std() Group UDF: Example 13
- 15. Group UDF: Example 15 80% of the code is boilerplate Slow
- 16. • Inefficient data movement between Java and Python (Serialization / Deserialization) • Scalar computation model UDF Issues 16
- 17. Apache Arrow 17
- 18. • In memory columnar format • Building on the success of Parquet • Standard from the start: • Developers from 13+ major open source projects involved • Benefits: • Share the effort • Create an ecosystem Apache Arrow 18 Calcite Cassandra Deeplearning4j Drill Hadoop Hbase Ibis Impala Kudu Pandas Parquet Phoenix Spark Storm R
- 19. High Performance Sharing & Interchange Before With Arrow
- 20. Columnar Data Format persons = [{ name: ’Joe', age: 18, phones: [ ‘555-111-1111’, ‘555-222-2222’ ] }, { name: ’Jack', age: 37, phones: [ ‘555-333-3333’ ] }]
- 21. Record Batch Construction Schema Dictionary Batch Record Batch Record Batch Record Batch name (offset) name (data) age (data) phones (list offset) phones (data) data header (describes offsets into data) name (bitmap) age (bitmap) phones (bitmap) phones (offset) { name: ’Joe', age: 18, phones: [ ‘555-111-1111’, ‘555-222-2222’ ] } Each box (vector) is contiguous memory The entire record batch is contiguous on wire
- 22. • Maximize CPU throughput • Pipelining • SIMD • Cache locality • Scatter/gather I/O In Memory Columnar Format for Speed
- 23. • PySpark “toPandas” Improvement • 53x Speedup • Streaming Arrow Performance • 7.75GB/s data movement • Arrow Parquet C++ Integration • 4GB/s reads • Pandas Integration • 9.71GB/s Results Read more on http://arrow.apache.org/blog/ 23
- 26. How PySpark UDF works 26 Executor Python Worker UDF: Row -> Row Rows (Pickle) Rows (Pickle)
- 27. • Inefficient data movement (Serialization / Deserialization) • Scalar computation model Recap: Current issues with UDF 27
- 28. Profile lambda x: x+1 8 Mb/s 91.8% in Ser/Deser
- 29. Vectorized UDF Executor Python Worker UDF: pd.DataFrame -> pd.DataFrame Rows -> RB RB -> Rows
- 30. Row UDF vs Vectorized UDF * Actual runtime for row UDF is 2s without profiling 20x Speed Up (Profiler overhead adjusted*)
- 31. Row UDF vs Vectorized UDF Ser/Deser Overhead Removed
- 32. Row UDF vs Vectorized UDF Less System Call Faster I/O
- 34. • Split-apply-combine • Break a problem into smaller pieces • Operate on each piece independently • Put all pieces back together • Common pattern supported in SQL, Spark, Pandas, R … Introduce Group UDF
- 35. • Split: groupBy • Apply: UDF (pd.DataFrame -> pd.DataFrame) • Combine: Inherently done by Spark Split-Apply-Combine (UDF)
- 36. Introduce groupBy().apply() Rows Rows Rows Groups Groups Groups Groups Groups Groups Each Group: pd.DataFrame -> pd.DataFramegroupBy
- 37. • (values – values.mean()) / values.std() Previous Example 37
- 38. Group UDF: Before and After For updated API, see: https://databricks.com/blog/2017/10/30/introducing-vectorized-udfs-for-pyspark.html Before: After*:
- 40. • Available in the upcoming Apache Spark 2.3 release • Try it with Databricks community version: • https://databricks.com/blog/2017/10/30/introducing-vectorized-udfs- for-pyspark.html Try It! 40
- 41. • Improving PySpark/Pandas interoperability (SPARK-22216) • Working towards Arrow 1.0 release • More Arrow integration Future Roadmap 41
- 42. • [email protected] • [email protected] Get involved 42
- 43. Bryan Cutler Hyukjin Kwon Jeff Reback Leif Walsh Li Jin Liang-Chi Hsieh Reynold Xin Takuya Ueshin Wenchen Fan Wes McKinney Xiao Li Collaborators 43
- 44. Questions 44