Nak alternatives?

[Eirik Rosvold Larsen]

Seems like Nak is no longer maintained on GitHub. It's a pity since it's a nice framework. It's now in a state where the most recent version of Nak is no longer binary compatible with the most recent version of Breeze (NoSuchMethodError etc)

I've been using NaiveBayes, TfIdf and BagOfWords classifiers. Are there any known good and maintained alternatives for Scala (preferably) or Java that does an equivalent job?

Regards, Eirik

[David Hall]

sorry about that :(

I tried to update it at some point but it wasn't terribly easy. Those particular classes may be easy to port.

I don't know of any good alternatives for "small/medium data" learning. MLLib for Spark is generally pretty solid, but is likely overkill.

-- David

[Jason Baldridge]

I may return to Nak at some point in the next few months and bring it back up to date. At my startup, I also wrote some code for training models with Vowpal Wabbit (defining features etc in Scala, dumping out a VW ready file, letting VW executable estimate the parameters, reading the parameters back into Scala. This has worked well for training lots of models with VW's speed and many nice features, while still being able to deploy models on the JVM. That might get open sourced soonish too.

[Eirik Rosvold Larsen]

Looking forward to nak update ;)

-Eirik

[Eirik Rosvold Larsen]

I'm pulling in NaiveBayes into my own codebase for now, in order to make it play with breeze 0.12.

-Eirik