Fork me on GitHub
#data-science
<
2017-09-29
>
shark8me00:09:14

You might want to look at another matrix library called core.matrix (https://github.com/mikera/core.matrix), and a neural network library called Cortex (https://github.com/thinktopic/cortex), which uses core.matrix for its number crunching operations.

rustam.gilaztdinov07:09:20

Thank you! Yes, I know about this libraries, but compare benchmarks, that's my point. And also, cortex, in my opinion, just yet another dl framework, nothing special. No new cool layers or preprocessing features

lepistane04:09:52

@rustam.gilaztdinov you can find more info about neanderthal http://dragan.rocks/ that's is the authors site, he usually writes blogs about it, just recently he wrote about mathematics that could be great direction to become aware of what you dont know you dont know. hope this helps

rustam.gilaztdinov07:09:38

That's exactly what i'm searching, many thx to you, @U45SLGVHV !

blueberry09:09:21

@rustam.gilaztdinov not to mention that the clojurecl test source contains all relevant parts of the book OpenCL in Action worked out in ClojureCL. I actually don't know of any Java library that has anything to do with GPU programming that has so mush documentation and code examples as Neanderthal+ClojureCL+ClojureCUDA (+ Bayadera)

blueberry09:09:12

Of course, implementing your own neural networks library might require some "non trivial" knowledge regardless of the library you use ;)

rustam.gilaztdinov11:09:58

Yes, it's true. Now I found your blog, @blueberry and that exactly answer to my first question. Thank you for sharing all of this, it's really great!! 💯 👏