13

I know that v10 has support for both Machine Learning and Finite Markov Processes, the Q-learning algorithm uses both. Specifically, Q-learning finds an optimal action-selection policy for any given (finite) Markov decision process (MDP).

Are there any specific combinations of functions and options that I should use as a starting point? or should I do this from scratch?

M.R.
  • 31,425
  • 8
  • 90
  • 281

1 Answers1

12

Right now you can use MDPtoolbox via RLink.

Reinforcement learning is expected in the next version of Mathematica. You can see this presentation from the Wolfram Conference 2015. A lot of interesting things are expected in the machine learning functionality (slides 10, 11 or images below).

enter image description here enter image description here

Alexey Golyshev
  • 9,526
  • 2
  • 27
  • 57