aria

8 ms·

1.
by aria 9y ago
Yes this does dynamic nets, the model is the same as PyTorch, you build a graph for each input(s). I think Cortex is in the Keras-style of model (only layer abstractions and not "dynamic" in the sense we talk about neural nets) an
2.
by aria 9y ago
I'll try! I've got a full time management job so this was a fun way to code again. The library is pretty small and I think easy to follow if you want to learn
3.
by aria 9y ago
Definitely doing the standard back-prop algorithm. Check out the code, it's pretty straightforward and the whole lib is about 2,500 lines
4.
by aria 9y ago
Author here, happy to answer any questions!
5.
by aria 9y ago
Author here. This is coming up. Will do OpenCL first and then CUDA.
6.
by aria 9y ago
Author here, happy to answer questions and hopefully convince someone to give the library a try.
7.
Flare: Clojure Dynamic Neural Net Library(aria42.com)
8 points by aria 9y ago|2 comments
8.
by aria 12y ago
The interesting part of BFGS is that it directly approximates H^{-1} rather than H.
9.
by aria 12y ago
Want to try a refresh, I think it ought to be fixed now. Thanks
10.
by aria 12y ago
In practice you only need to do $H^{-1} g$; L-BFGS stores the {s_k} and {y_k} vectors which allow you to do the $H^{-1} g$ directly rather than needing to ever form the hessian or its inverse. There are techniques that aren't BFGS-base
11.
by aria 12y ago
Typo is fixed, thanks! I think that's explicitly mentioned in the Quasi-Newton section that you only need to implicitly multiply and not form the matrix.
12.
by aria 12y ago
What browser are you in? It just using MathJax and appears to work in WebKit and Firefox.
13.
by aria 12y ago
Author here, happy to answer any questions.
14.
Numerical Optimization: Understanding L-BFGS(aria42.com)
34 points by aria 12y ago|14 comments
15.
by aria 13y ago
Question 1: What is the first bit in your unique 33-bit string? Question 2: What is the first bit in your unique 33-bit string? ...
16.
by aria 13y ago
Author here. Happy to answer any questions!
17.
Clojure Community Night(blog.getprismatic.com)
26 points by aria 13y ago|5 comments
18.
by aria 13y ago
Thanks! I definitely have personally run into the issue a lot and our Clojure teams at Prismatic struggled with commenting discipline. We find schemas are easier to maintain and apply, plus they save a lot of time during dev and testing.
19.
by aria 13y ago
Thanks! Yeah I think a traditional type system wouldn't make sense for Clojure, you need something that can describe data structures more succinctly, which is what we were shooting for
20.
by aria 13y ago
Definitely an inspiration. The core utility of Schema is that the declarations are still data and can be utilized in code. I think we'll have some interesting applications of this idea soon...
21.
by aria 13y ago
Thanks so much. We spent a long time thinking about how to have the benefits of types in a Clojure-y data-oriented way. The advantage of having Schemas as data is that we can use Clojure code to process them and generate things like Object
22.
by aria 13y ago
Author here. Happy to answer any questions or comments.
23.
by aria 13y ago
Author here. Happy to answer any questions!
24.
by aria 13y ago
I could be wrong, but I don't think Incanter has any Clojure-native means of generic operations over arrays at all.
25.
by aria 13y ago
Have any icon ideas? Was thinking ["hip","hip"] or hiphip[]
26.
by aria 13y ago
Indeed, the core logic of HipHip is the same for all primitive types and macros generate type-hinted versions for each primitive type.
27.
by aria 13y ago
Definitely not the same thing. Take a look at our earlier graph post. http://blog.getprismatic.com/blog/2013/2/1/graph-abstraction...
28.
by aria 13y ago
Graph would be possible, but not as clean in Python. Consider switching up drugs. Clojure is way more potent man.
29.
by aria 13y ago
We internally have a nice component/widget abstraction we use which is not quite MVC, but we think cleans up a lot of web development. We will release in the near future once we perfect it.
30.
by aria 13y ago
On mobile devices, we were hitting significant performance barriers from DOM and class manipulation. YMMV but it seems DOM performance is an issue on non-beefy boxes.

More ›