Does anyone know of any links on the Wolfram site that could be used to benchmark implementing a reinforcement learning application in Mathematica vs. Wolfram System Modeler? There was one Mathematica Reinforcement Learning Notebook example but I am hoping to find a more complete example for each environment (Notebook & System Modeler) to help drive the decision on which to use to for our application.
Does anyone know of a good set of reference examples?
Could you elaborate on the problem you want to solve? (for example, what is the environment you want to train an agent on?) SystemModeler is appropriate for building complex environments that you want to train agents on (but I don't think you can use SystemModeler to train agents).
There is also this talk that you might find useful, showing how to train an agent on environments defined in SystemModeler and OpenAI Gym: http://www.wolfram.com/broadcast/video.php?c=104&p=2&v=2139
The environment in which we want to train the agent is a control system where the the current state of the environment is identified by a classifier which quantizes a continuous range of environment states into a smaller set of good and bad states. So our problem is similar to the common pole balancing example with some quantization of the environment states using a CNN.
Your comment on using SystemModeler for building the complex environment but not to train the agents is exactly what I was interested in understanding. Based on this comment it appears that we may be able to stay within the Mathematica framework for our implementation if we can adequately represent our environment with a CNN. Do you agree?
That is correct.
Note also: we should have some reinforcement learning examples in the upcoming 12 release documentation. We are busy adding a simple feature to the neural net framework that will allow you train agents using using the recently popular policy-gradient methods.