Fork me on GitHub
#numerical-computing
<
2017-03-25
>
qqq06:03:10

active channel here 🙂

qqq06:03:14

so my choices right now are:

qqq06:03:53

(1) use Hy + tensorflow (2) try to get Tensorflow 1.0's compleetly undocumented java api to work (3) use Clojure + Cortex (4) use Clojure + neathenaderal (only provides blas) (5) use osmething else

qqq06:03:56

suggestions ?

blueberry08:03:40

@qqq A few comments regarding option 4, although you didn't write what you are trying to do:

blueberry08:03:24

1) Neanderthal also supports LAPACK from the upcoming version 0.9.0 that is going to be released in a few days.

blueberry08:03:04

2) Neanderthal also supports straightforward communication with GPUs (OpenCL based, though)

blueberry08:03:38

3) I am going to create ClojureCUDA soon (a couple of months I guess)

blueberry08:03:21

4) In one of the next versions, Neanderthal will also have CUDA engine and easy transfer to CUDA

blueberry08:03:50

5) Then, cuDNN is a (relatively) straightforward library away...

qqq17:03:06

@blueberry: sorry, my XY problem is: I am doing deep learning in Hy/Tensorflow

qqq17:03:20

I'm sick and tired of Hy being sorta clojure but not really clojure and want my entire stack in clojure.

qqq17:03:42

I'm willing to give up numpy and it's entire ecosystem -- I use mostly tensorflow, not so much numpy