|
|
## Scheduling
|
|
|
|
|
|
An approach for finding optimal schedulers via reinforcement learning in nondeterministic Hybrid Petri nets with general transitions is implemented on the schedulingML branch and for models including non-linear ODES on dynamicapprox.
|
|
|
|
|
|
Model checking properties can be maximized and minimized (usage similar to **check** but with keyword **maximize**/**minimize**)
|
|
|
|
|
|
New parameters are introduced which can be changed by **change _parametername_**
|
|
|
- _parametername_ can be:
|
|
|
+ **trainingruns**: amount of trainingruns
|
|
|
+ **discretization**: discretization of the continuous variables for the scheduler, 10 truncates after the first decimal place, 100 after the second etc.
|
|
|
+ **prophetic**: true when the scheduler has information on the future general transition firings
|
|
|
+ **discount factor**: discount factor parameter required for training
|
|
|
+ **alpha**: alpha parameter required for training
|
|
|
+ **epsilon**: epsilon parameter for epsilon-greedy exploration
|
|
|
- Mandatory option: **--n** _newvalue_: Number or decimal/boolean with the new parameter value. |
|
|
\ No newline at end of file |