Fork me on GitHub
#core-matrix
<
2015-07-29
>
mjs260015:07:23

Has anyone used weka with clojure? If so, how painful was it?

shriphani20:07:09

@mjs2600: Weka is pretty shit.

shriphani20:07:23

The gold standard for a well-maintained ML library is still scikit.

shriphani20:07:55

I would recommend doing your modeling outside the jvm and then moving it to the jvm via a CSV file or somesuch

mjs260020:07:01

@shriphani: I’m doing reinforcement learning, so scikit-learn is out (that’s my normal approach).

shriphani20:07:25

if this is plain Q-learning, why not just implement the bellman eq in clojure ?

mjs260020:07:12

I’m taking the OMSCS class on RL. I actually did exactly that in python for ML.

mjs260020:07:38

It looks like BURLAP is the most complete RL library, but it doesn’t look great...

shriphani20:07:43

ah. not familiar with that.

shriphani20:07:52

but I would recommend rolling your own

shriphani20:07:55

how complex is the model ?

mjs260020:07:59

It’s built on weka, which is why I asked about that.

mjs260020:07:12

None of them are going to be terrible, but if it was simple to work with, I didn’t want to rewrite a whole library.

mjs260020:07:46

Thanks for the heads up, I guess I’ll write my own. Do you know if clojure is at all competitive with numpy for matrix stuff?

mjs260020:07:23

I’ve done a bunch of work in clojure but all of my ML stuff has been in python.