Learning Stochastic Systems

Active stochastic learning follows the same three step procedure as deterministic and non-deterministic learning.

Only differance is that you need to invoke run_stochastic_Lstar. Aside from input alphabet, SUL implementation, and equivalence oracle, you also need to provide automaton_type value. It can either be mdp or smm, for learning Markov decision processes or Stochastic Mealy machines respectively. Note that in most, but not all cases, smm is preferred as its representation is more compact and will learn faster.

Note that the equivalence oracle has to be either Random Walk or Random Word oracle.

If you are interested in the algorithm itself, check out L* Based Learning of MDPs and we will provide link to the Stochastic Mealy Machine learning paper, that contains other optimizations, in near feature.