internal static MarkovDecisionProcess Create()
{
// Just a simple MDP with simple nondeterministic choices
// ⟳0⟶1⟲
var mdp = new MarkovDecisionProcess(ModelCapacityByMemorySize.Tiny);
mdp.StateFormulaLabels = new string[] { Label1Formula.Label, Label2Formula.Label };
mdp.StateRewardRetrieverLabels = new string[] { };
mdp.StartWithInitialDistributions();
mdp.StartWithNewInitialDistribution();
mdp.AddTransitionToInitialDistribution(0, 1.0);
mdp.FinishInitialDistribution();
mdp.FinishInitialDistributions();
mdp.SetStateLabeling(1, new StateFormulaSet(new[] { true, false }));
mdp.StartWithNewDistributions(1);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(0, new StateFormulaSet(new[] { false, true }));
mdp.StartWithNewDistributions(0);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 1.0);
mdp.FinishDistribution();
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
return(mdp);
}
internal static MarkovDecisionProcess Create()
{
// A MDP which was designed to test prob1e
// 0⟶0.5⟼1⟲ 0.5⇢0
// 0.5⟼3⟶4➞0.5⟲
// ↘2⟲
var mdp = new MarkovDecisionProcess(ModelCapacityByMemorySize.Tiny);
mdp.StateFormulaLabels = new string[] { Label1Formula.Label, Label2Formula.Label };
mdp.StateRewardRetrieverLabels = new string[] { };
mdp.StartWithInitialDistributions();
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 1.0);
mdp.FinishDistribution();
mdp.FinishInitialDistributions();
mdp.SetStateLabeling(0, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(0);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 0.5);
mdp.AddTransition(3, 0.5);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(1, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(1);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(2, new StateFormulaSet(new[] { true, false }));
mdp.StartWithNewDistributions(2);
mdp.StartWithNewDistribution();
mdp.AddTransition(2, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(3, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(3);
mdp.StartWithNewDistribution();
mdp.AddTransition(2, 1.0);
mdp.FinishDistribution();
mdp.StartWithNewDistribution();
mdp.AddTransition(4, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(4, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(4);
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 0.5);
mdp.AddTransition(4, 0.5);
mdp.FinishDistribution();
mdp.FinishDistributions();
return(mdp);
}
private void AddPaddingStatesInMdp()
{
foreach (var paddingState in _paddingStates)
{
if (paddingState.Key.IsInitial)
{
MarkovDecisionProcess.StartWithInitialDistributions();
}
else
{
MarkovDecisionProcess.StartWithNewDistributions(paddingState.Key.State);
}
MarkovDecisionProcess.StartWithNewDistribution();
MarkovDecisionProcess.AddTransition(paddingState.Value, 1.0);
MarkovDecisionProcess.FinishDistribution();
if (paddingState.Key.IsInitial)
{
MarkovDecisionProcess.FinishInitialDistributions();
}
else
{
MarkovDecisionProcess.FinishDistributions();
}
}
}
internal static MarkovDecisionProcess Create()
{
// MDP of [Parker02, page 36]
// 0
// ⇅
// 1➞0.6⟼2⟲
// 0.4⟼3⟲
var mdp = new MarkovDecisionProcess(ModelCapacityByMemorySize.Tiny);
mdp.StateFormulaLabels = new string[] { Label1Formula.Label, Label2Formula.Label };
mdp.StateRewardRetrieverLabels = new string[] { };
mdp.StartWithInitialDistributions();
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 1.0);
mdp.FinishDistribution();
mdp.FinishInitialDistributions();
mdp.SetStateLabeling(0, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(0);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(1, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(1);
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 1.0);
mdp.FinishDistribution();
mdp.StartWithNewDistribution();
mdp.AddTransition(2, 0.6);
mdp.AddTransition(3, 0.4);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(2, new StateFormulaSet(new[] { true, false }));
mdp.StartWithNewDistributions(2);
mdp.StartWithNewDistribution();
mdp.AddTransition(2, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(3, new StateFormulaSet(new[] { false, true }));
mdp.StartWithNewDistributions(3);
mdp.StartWithNewDistribution();
mdp.AddTransition(3, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
return(mdp);
}
private void AddDistribution(int distribution)
{
if (_ltmdpContinuationDistributionMapper.IsDistributionEmpty(distribution))
{
return;
}
MarkovDecisionProcess.StartWithNewDistribution();
var enumerator = _ltmdpContinuationDistributionMapper.GetContinuationsOfDistributionEnumerator(distribution);
while (enumerator.MoveNext())
{
var leaf = GetLeafOfCid(enumerator.CurrentContinuationId);
var probability = GetProbabilityOfCid(enumerator.CurrentContinuationId);
MarkovDecisionProcess.AddTransition(leaf.ToState, probability);
}
MarkovDecisionProcess.FinishDistribution();
}
private void AddPaddedTransition(int mdpState, double probability, int requiredPadding)
{
var firstStateBeforePadding = CreateNewArtificialPaddingStatesBackward(mdpState, requiredPadding);
MarkovDecisionProcess.AddTransition(firstStateBeforePadding, probability);
}
