GPU Implementation?

[haskarb]

How can I use this package with Keras in optimiser class?
Any thoughts on GPU implementation?

[ljvmiranda921]

Hi, I've realized you're the same person in Issue #19

As of now, implementing this package on GPU (perhaps via CUDA) is not yet being considered in the Development Roadmap for v.0.2.0 (#5). In all honesty I don't have any experience in CUDA 😞 . If you can help and show us a lead, then perhaps we can add this feature for v.0.3.0.

For your first question, what you can do is take the update_* and optimize methods of a PSO optimizer from pyswarms, and use these methods to populate a custom-class that inherits from Keras' Optimizer base class. 👍

[haskarb]

@ljvmiranda921 I have written a GPU PSO code with pyCUDA. I will share with you once you start v.0.3.0.
Thanks.

[ljvmiranda921]

Awesome! Okay, I will ping you once we've started tackling this feature. Thanks a lot!

[ljvmiranda921]

Hmmm, I just realized that maybe I can leverage Pytorch (as a substitute to numpy) to take advantage of GPU computations. This is just an idea in the clouds but I'll cook something up this winter vacation.