maxrobinsonthegreat / neuralpatterns Goto Github PK

i've been super fascinated with these for quite some time now. i'm wondering if you have any tips for finding kernels and activation functions that have desired emergent features? or is it purely intuition?

It appears that every time the environment is restarted an additional 4 MB of memory is allocated without cleaning up the previous simulation.

Here's a snapshot of memory usage for the tab itself to demonstrate how easy it is to end up with GBs being allocated when restarting the environment a few times.

Due to the fact that fixing the leak itself may not be an option due to it being a problem with WebGL I theorize that a somewhat hacky fix is to rewrite Utils.generateState() by using the same logic the brush uses to clear the environment and to place the cells using that to recreate the same effect without having to make a whole new buffer.

Hi,

I've been spending a ton of time and having lots of fun on neuralpatterns.io. Awesome project! But I noticed I could not use bitwise operators, not even bitwise AND:

ERROR: 0:56: '&' : bit-wise operator supported in GLSL ES 3.00 and above

I tried the directive #version 300 es but I got ERROR: 0:21: 'version' : #version directive must occur before anything else, except for comments and white space. I'm guessing the Activation code block is getting injected into part of a greater runtime.

I know next to nothing about GLSL so I have no idea if this is feasible with the runtime you are using or not. But it would be super awesome if the default shader used a slightly newer version. Maybe you could even add a field which lets a user pass preprocessor directives to compileShaders.