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 prob0e
// 4
// ⇅
// 0⟼1↘
// ↘2⟼3⟲
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(0, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(0);
mdp.StartWithNewDistribution();
mdp.AddTransition(1, 1.0);
mdp.FinishDistribution();
mdp.StartWithNewDistribution();
mdp.AddTransition(2, 1.0);
mdp.FinishDistribution();
mdp.StartWithNewDistribution();
mdp.AddTransition(4, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(1, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(1);
mdp.StartWithNewDistribution();
mdp.AddTransition(3, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(2, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(2);
mdp.StartWithNewDistribution();
mdp.AddTransition(3, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(3, new StateFormulaSet(new[] { true, false }));
mdp.StartWithNewDistributions(3);
mdp.StartWithNewDistribution();
mdp.AddTransition(3, 1.0);
mdp.FinishDistribution();
mdp.FinishDistributions();
mdp.SetStateLabeling(4, new StateFormulaSet(new[] { false, false }));
mdp.StartWithNewDistributions(4);
mdp.StartWithNewDistribution();
mdp.AddTransition(0, 1.0);
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.StartWithNewInitialDistribution();
mdp.AddTransitionToInitialDistribution(0, 1.0);
mdp.FinishInitialDistribution();
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 FinishDistributions(int?stateToStartFrom)
{
if (stateToStartFrom.HasValue)
{
MarkovDecisionProcess.FinishDistributions();
}
else
{
MarkovDecisionProcess.FinishInitialDistributions();
}
}
private void ConvertChildCid(long currentCid)
{
var distanceFromRootOfSourceNode = _cidDistanceFromRoot[currentCid];
var cge = _nmdp.GetContinuationGraphElement(currentCid);
if (cge.IsChoiceTypeUnsplitOrFinal || cge.IsChoiceTypeForward)
{
return;
}
var mdpState = _cidToArtificialStateMapping[currentCid];
MarkovDecisionProcess.StartWithNewDistributions(mdpState);
var cgi = _nmdp.GetContinuationGraphInnerNode(currentCid);
if (cge.IsChoiceTypeNondeterministic)
{
for (var i = cgi.FromCid; i <= cgi.ToCid; i++)
{
MarkovDecisionProcess.StartWithNewDistribution();
AddDestination(i, distanceFromRootOfSourceNode);
MarkovDecisionProcess.FinishDistribution();
}
}
else if (cge.IsChoiceTypeProbabilitstic)
{
MarkovDecisionProcess.StartWithNewDistribution();
for (var i = cgi.FromCid; i <= cgi.ToCid; i++)
{
AddDestination(i, distanceFromRootOfSourceNode);
}
MarkovDecisionProcess.FinishDistribution();
}
MarkovDecisionProcess.FinishDistributions();
for (var i = cgi.FromCid; i <= cgi.ToCid; i++)
{
ConvertChildCid(i);
}
}
private void ConvertStateTransitions()
{
for (var state = 0; state < _nmdp.States; state++)
{
var cidOfStateRoot = _nmdp.GetRootContinuationGraphLocationOfState(state);
Clear(cidOfStateRoot);
_ltmdpContinuationDistributionMapper.AddInitialDistributionAndContinuation(cidOfStateRoot);
UpdateContinuationDistributionMapperAndCollectLeafs(cidOfStateRoot);
MarkovDecisionProcess.StartWithNewDistributions(state);
var numberOfDistributions = _ltmdpContinuationDistributionMapper.GetNumbersOfDistributions();
for (var distribution = 0; distribution < numberOfDistributions; distribution++)
{
AddDistribution(distribution);
}
MarkovDecisionProcess.FinishDistributions();
}
}
