Functional Programming for Machine Learning and Data Pipelines

on YouTube ↗40m

" summary

Argument: functional programming plus a strong type system reduces implementation bugs and makes ML code reusable. Concrete tools shown: Scala type classes and a custom data type class to write algorithms that work on Scala collections and Spark RDDs; Breeze/Spire for numeric vectors. Demonstrations: implementations of k-means and k-nearest neighbors using sparse feature vectors on the 20 Newsgroups set; a live kNN classifier runs but yields only ~12% accuracy in the quick demo. Code and slides are linked from the uploader.

AI-written summary · about

" chapters (8)

1. 0:00Introduction and motivation
2. 3:31Why correctness matters
3. 7:29Functional programming primer
4. 11:50Machine learning basics
5. 18:52Type systems in ML
6. 23:11Type classes for ML
7. 29:37Classic ML problems
8. 37:23Live demo and conclusion

" related talks

• 59mFunctional Programming for Machine Learning PanelVitaly Gordon, Malcolm Greaves · Scale By the Bay 2017
• 21mFP for Data Science For Loops Considered HarmfulAris Vlasakakis · Scale By The Bay 2018
• 43mFunctional Big DataChris Richardson · Great Indian Developer Summit 2015
• 57mThe Fun of Functional Programming PanelJohn A De Goes, Bill Venners, Julie Pitt, Dick Wall, Michael Pilquist, Stew O'Connor · Scale By the Bay 2017
• 44mFunctional Programming You Already KnowKevlin Henney · Codemotion Rome 2015
• 28mFunctional Linear Algebra in ScalaVlad Patryshev · Scale By the Bay 2017
• 47mPlain Functional ProgrammingMartin Odersky · Devoxx Belgium 2017
• 47mKeeping the fun in Apache Spark Datasets and FPHolden Karau · Scala Days Berlin 2018