Functional Programming & Scala: The Killer Combo for Machine Learning • Marek Kolodziej • YOW! 2015

From the creator

This presentation was recorded at YOW! 2015. #GOTOcon #YOW https://yowcon.com Marek Kolodziej - Principal Research Engineer at Nitro, Inc. @mkolod RESOURCES https://x.com/marekinfo https://github.com/mkolod https://www.linkedin.com/in/marek-kolodziej ABSTRACT While FP and Scala have already become the mainstays of middleware, web development and big data stacks (#Akka, Play, Kafka, Spark), they tend not to have a big presence in the machine learning and NLP communities. For instance, the emerging deep learning toolkits are mostly Python‐based (#Pylearn2, Theano, etc.). The same goes for general-purpose machine learning (Python's scikit-learn, countless R libraries). Performance seekers dissatisfied with slow scripting languages write typed Cython code, contorted C++ libraries bound to scripting language wrappers, or resort to random exotic solutions such as Lua. Some even dispense with all abstraction and write incomprehensible CUDA kernels. There has to be a better way. As a machine learning engineer, I want to write strongly typed functional code. Math has no place for side effects, and I don't want to waste time running a simulation for hours, only to find that I made a typo in my "stringly-typed" script. Unbeknownst to most, Scala's machine learning and NLP ecosystem is growing rapidly, from numeric processing (Spire, Breeze) to big data machine learning (MLLib, Mahout) to GPU‐based text parsing (Puck), to general‐purpose probabilistic programming (FACTORIE). In thi

Published 2024-09-01 · Watch on YouTube ↗
Availability and playback restrictions are controlled by the original channel.