Re:EXGBoost: Gradient Boosting in Elixir
by mrdoops on Wednesday July 26, 2023 @04:20PM
(#36884361)
Elixir is good at doing a lot of things at once on - scaling to lots of machines - and not exploding catastrophically while doing so.
Turns out this is really helpful for machine learning where you want to coordinate big data pipelines and do things like batching requests to a GPU resource (because GPUs want to be parallelized).
Re:EXGBoost: Gradient Boosting in Elixir
by impulser_ on Wednesday July 26, 2023 @02:35PM
(#36882676)
Elixir has always been a well liked language on here and in general. The past year there has been a lot of work on building data and AI/ML libraries in Elixir.
Explorer, Nx, Axon, Scholar, Bumblebee, Broadway, Livebook and now EXGBoost.
You can get pretty close to building a full stack AI application in Elixir now.
From data ingestion and processing all the way to serving the model in a Phoenix web app. All in Elixir.
I think it's great more languages are becoming alternates to Python in data and AI/ML, more options the better.
Re:EXGBoost: Gradient Boosting in Elixir
by lawik on Wednesday July 26, 2023 @02:40PM
(#36882760)
There have been a few just recently. Bit of an ML boom in Elixir and ML is popular right now.
As an Elixir-centric blogger I can tell you that Elixir generally seems to do well on Hacker News. I think it has a lot of overlap with interests on here. Erlang is the underpinnings and Erlang is generally interesting from a comp sci standpoint. Elixir's roots in Ruby line it up nicely with SaaS, startups and that whole space.
I am only reposting my own tweet here because I just sent an answer to the exact same question to a youtuber who was curious: https://twitter.com/pmarreck/status/1684248660832288771?s=20
It has a few links.
There's also this podcast (which has a transcript link): https://smartlogic.io/podcast/elixir-wizards/s10-e10-sean-mo...
Turns out this is really helpful for machine learning where you want to coordinate big data pipelines and do things like batching requests to a GPU resource (because GPUs want to be parallelized).
You can do batched ML inference pretty much out of the box with Nx.Serving https://hexdocs.pm/nx/Nx.Serving.html where you'd have to spin up a separate third party service like https://developer.nvidia.com/triton-inference-server otherwise.
Explorer, Nx, Axon, Scholar, Bumblebee, Broadway, Livebook and now EXGBoost.
You can get pretty close to building a full stack AI application in Elixir now.
From data ingestion and processing all the way to serving the model in a Phoenix web app. All in Elixir.
I think it's great more languages are becoming alternates to Python in data and AI/ML, more options the better.
As an Elixir-centric blogger I can tell you that Elixir generally seems to do well on Hacker News. I think it has a lot of overlap with interests on here. Erlang is the underpinnings and Erlang is generally interesting from a comp sci standpoint. Elixir's roots in Ruby line it up nicely with SaaS, startups and that whole space.