Various algorithms?

[ericvrp]

Will this repo be the place where multiple reinforcement learning algorithms (q-learning, A3C, ...) be implemented for Keras?

[farizrahman4u]

There is no successful implementation of A3C in Keras yet. I will work on it if there is enough demand. I wrote this project with simplicity in mind, so that deep learning enthusiasts could test various models without bothering about complications that comes with game simulators etc.. Adding your own RL algorithm or game is pretty straight forward, just go through the code for existing agent / games.

I also suggest you take a look at https://github.com/osh/kerlym.

[srinivasjdi]

@farizrahman4u : For A3C, I'm using the Tensorflow implementation from async_deep_reinforce. But, for Q-learning I use qlearning4k. It would be definitely useful to write A3C in keras or Tflearn. I'ld be willing to join in for the A3C implementation. Also, one feature request for qlearning4k is to add a pygame visualization screen. You already have what you need in the 'canvas' variable. It's just a matter of displaying it ( at the right scale and colour) on pygame window.

[farizrahman4u]

@srinivasjdi For pygame integration, you can checkout www.github.com/farizrahman4u/KerasPlaysFlappyBird

[ghost]

https://github.com/deepersystems/bg