How to use GA in Reinforcement Learning instead of Gradient descent?

I made my Environment (reset and step functions) but I want to use Genetic algorithm for NN tuning instead of gradient descent. How to do this?

Answer 1

[Emmanouil Tzorakoleftherakis]

Hello,

Evolutionary RL is not provided out of the box as of now. To use it you would have to implement a custom training loop (see here for an example).

[Alexander]

I made custom training loop. I see that actorNetwork has several layers (including FullyConnected) but their Weights are [ ]. Where are weights of actorNetwork?

[Emmanouil Tzorakoleftherakis]

Please refer to this page on how to extract the weights of actors/critics.