public override void UpdateMemoryBlocks()
{
UtilityOutput.Count = NoActions;
RewardStats.Count = 2;
LearningParams = new MyModuleParams();
Memory = new MyQSAMemory(GlobalDataInput.Count, NoActions);
LearningAlgorithm = new MyDiscreteQLearning(LearningParams, Memory);
if (GlobalDataInput != null)
{
if (NoActions == 6)
{
MyLog.DEBUG.WriteLine("6 actions set by the user, will use action names for gridworld");
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, new String[] { " -", " <", " >", " ^", " v", " P" }, LearningParams);
}
else if (NoActions == 3)
{
MyLog.DEBUG.WriteLine("3 actions set by the user, will use action names for pong");
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, new String[] { " <", " -", " >" }, LearningParams);
}
else
{
MyLog.DEBUG.WriteLine("Unknown no. of actions, will use automatic naming of actions");
String[] names = new String[NoActions];
for (int i = 0; i < NoActions; i++)
{
names[i] = "A" + i;
}
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, names, LearningParams);
}
CurrentStateOutput.Count = GlobalDataInput.Count;
}
}
public MyActionHistory(ActionManager am, MyModuleParams setup)
{
this.setup = setup;
this.am = am;
this.changed = new List <int>();
this.isMonitored = new Dictionary <int, bool>();
}
public override void UpdateMemoryBlocks()
{
UtilityOutput.Count = NoActions;
LearningParams = new MyModuleParams();
if (GlobalDataInput != null)
{
if (NoActions == 6)
{
MyLog.DEBUG.WriteLine("6 actions set by the user, will use action names for gridworld");
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, new String[] { " -", " <", " >", " ^", " v", " P" }, LearningParams);
}
else if (NoActions == 3)
{
MyLog.DEBUG.WriteLine("3 actions set by the user, will use action names for pong");
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, new String[] { " <", " -", " >" }, LearningParams);
}
else
{
MyLog.DEBUG.WriteLine("Unknown no. of actions, will use automatic naming of actions");
String[] names = new String[NoActions];
for (int i = 0; i < NoActions; i++)
{
names[i] = "A" + i;
}
Rds = new MyRootDecisionSpace(GlobalDataInput.Count, names, LearningParams);
}
Hierarchy = new MyHierarchyMaintainer(Rds, LearningParams);
SubspaceUtils.Count = MaxActions * MaxActions;
SubspaceUtils.ColumnHint = MaxActions;
}
}
public MyActionHistory(ActionManager am, MyModuleParams setup)
{
this.setup = setup;
this.am = am;
this.changed = new List<int>();
this.isMonitored = new Dictionary<int, bool>();
}
public MyDiscreteQLearning(MyModuleParams learningParams, MyQSAMemory mem)
{
this.m_mem = mem;
this.m_learning = learningParams;
m_trace = new MyEligibilityTrace(learningParams);
m_trace.PushState(new int[mem.GetMaxStateVariables()]);
}
public MyDiscreteQLearning(MyModuleParams learningParams, MyQSAMemory mem)
{
this.m_mem = mem;
this.m_learning = learningParams;
m_trace = new MyEligibilityTrace(learningParams);
m_trace.PushState(new int[mem.GetMaxStateVariables()]);
}
public MyVariableHistory(VariableManager vm, MyModuleParams setup)
{
this.vm = vm;
this.setup = setup;
this.changed = new List <List <int> >();
isMonitored = new bool[vm.GetMaxVariables()];
}
public MyDecisionSpace(MyStochasticReturnPredictor mySRP, MyRootDecisionSpace rds, MyModuleParams setup)
{
this.m_setup = setup;
this.m_rds = rds;
this.m_mySRP = mySRP;
m_S_t = new int[rds.VarManager.MAX_VARIABLES];
ChildActions = new MyActionLink(0, rds);
m_childVariables = new MyVariableLink(rds.VarManager.GetMaxVariables());
}
public ActionManager(String[] labels, MyModuleParams setup)
{
this.m_setup = setup;
Actions = new List<MyMotivatedAction>();
for (int i = 0; i < labels.Length; i++)
{
Actions.Add(new MyMotivatedAction(labels[i], 0, setup));
}
this.AcitonHistory = new MyActionHistory(this, setup);
this.AcitonHistory.AddAllCurrentActions();
}
public ActionManager(int noPrimitiveActions, MyModuleParams setup)
{
this.m_setup = setup;
Actions = new List <MyMotivatedAction>();
for (int i = 0; i < noPrimitiveActions; i++)
{
Actions.Add(new MyMotivatedAction("" + i, 0, setup));
}
this.AcitonHistory = new MyActionHistory(this, setup);
this.AcitonHistory.AddAllCurrentActions();
}
public ActionManager(int noPrimitiveActions, MyModuleParams setup)
{
this.m_setup = setup;
Actions = new List<MyMotivatedAction>();
for (int i = 0; i < noPrimitiveActions; i++)
{
Actions.Add(new MyMotivatedAction("" + i, 0, setup));
}
this.AcitonHistory = new MyActionHistory(this, setup);
this.AcitonHistory.AddAllCurrentActions();
}
public ActionManager(String[] labels, MyModuleParams setup)
{
this.m_setup = setup;
Actions = new List <MyMotivatedAction>();
for (int i = 0; i < labels.Length; i++)
{
Actions.Add(new MyMotivatedAction(labels[i], 0, setup));
}
this.AcitonHistory = new MyActionHistory(this, setup);
this.AcitonHistory.AddAllCurrentActions();
}
public MyAbstractHierarchy(MyRootDecisionSpace rds, MyModuleParams learningParams)
{
this.m_learningParams = learningParams;
m_actionLevels = new List <List <MyAction> >();
this.m_rds = rds;
for (int i = 0; i < rds.ActionManager.GetNoActions(); i++)
{
this.AddAction(i);
}
}
public MyLocalVariableHistory(MyRootDecisionSpace rds,
MyModuleParams setup, IDecisionSpace ds)
{
m_rds = rds;
m_setup = setup;
changes = new Dictionary <int, float>();
// all potential variables added by default
for (int i = 0; i < rds.VarManager.GetMaxVariables(); i++)
{
changes.Add(i, INIT_VAL);
}
}
public MyEligibilityTrace(MyModuleParams learning)
{
this.m_learning = learning;
m_traceData = new List <TraceData>();
Lambda = new float[learning.TraceLength];
float tmp = learning.Lambda;
for (int i = 0; i < learning.TraceLength; i++)
{
Lambda[i] = tmp;
tmp *= learning.Lambda;
}
}
public MyStochasticReturnPredictor(MyRootDecisionSpace rds, int myPromotedVariable,
MyModuleParams setup, String label, int level)
: base(label, level, setup)
{
base.AddPromotedVariable(myPromotedVariable, rds);
this.Rds = rds;
Ds = new MyDecisionSpace(this, rds, setup);
Mem = new MyQSAMemory(rds.VarManager.MAX_VARIABLES, 0);
m_asm = new MyMotivationBasedDeleteUnselectedASM(setup);
LearningAlgorithm = new MyDiscreteQLearning(setup, Mem);
m_mlvh = new MyLocalVariableHistory(rds, m_setup, Ds);
m_prevSelectedAction = 0;
m_prev_st = Ds.GetCurrentState();
this.m_newVariables = new List <int>();
}
private int[] m_prev_st; // previous state (for variable adding and sharing knowledge)
#endregion Fields
#region Constructors
public MyStochasticReturnPredictor(MyRootDecisionSpace rds, int myPromotedVariable,
MyModuleParams setup, String label, int level)
: base(label, level, setup)
{
base.AddPromotedVariable(myPromotedVariable, rds);
this.Rds = rds;
Ds = new MyDecisionSpace(this, rds, setup);
Mem = new MyQSAMemory(rds.VarManager.MAX_VARIABLES, 0);
m_asm = new MyMotivationBasedDeleteUnselectedASM(setup);
LearningAlgorithm = new MyDiscreteQLearning(setup, Mem);
m_mlvh = new MyLocalVariableHistory(rds, m_setup, Ds);
m_prevSelectedAction = 0;
m_prev_st = Ds.GetCurrentState();
this.m_newVariables = new List<int>();
}
public VariableManager(String[] labels, MyModuleParams setup)
{
this.m_labels = labels;
m_vars = new MyVariable[labels.Length];
if (MAX_VARIABLES <= 0)
{
MAX_VARIABLES = 1;
}
else
{
this.MAX_VARIABLES = labels.Length;
}
this.InitVars();
this.VarHistory = new MyVariableHistory(this, setup);
this.VarHistory.AddAllPotentialVariables();
}
public VariableManager(int maxVariables, MyModuleParams setup)
{
if (maxVariables <= 0)
{
maxVariables = 1;
}
else
{
this.MAX_VARIABLES = maxVariables;
}
m_labels = new String[this.MAX_VARIABLES];
for (int i = 0; i < m_labels.Length; i++)
{
m_labels[i] = "" + i;
}
this.InitVars();
this.VarHistory = new MyVariableHistory(this, setup);
this.VarHistory.AddAllPotentialVariables();
this.ShouldBeSubspaced = new List <int>();
}
public MyMotivationBasedDeleteUnselectedASM(MyModuleParams setup)
{
this.m_setup = setup;
m_rnd = new Random();
}
public MyMotivatedAction(String label, int level, MyModuleParams setup)
: base(label, level)
{
this.m_motivatonSource = new MyMotivationSource(setup, level);
this.m_setup = setup;
}
public MyRootDecisionSpace(int maxVariables, String[] actionNames, MyModuleParams setup)
{
VarManager = new VariableManager(maxVariables, setup);
ActionManager = new ActionManager(actionNames, setup);
}
public MyMotivatedAction(String label, int level, MyModuleParams setup)
: base(label, level)
{
this.m_motivatonSource = new MyMotivationSource(setup, level);
this.m_setup = setup;
}
public MyMotivationSource(MyModuleParams setup, int level)
{
this.m_setup = setup;
this.m_val = 0;
this.m_level = level;
}
public MyRootDecisionSpace(int maxVariables, String[] actionNames, MyModuleParams setup)
{
VarManager = new VariableManager(maxVariables, setup);
ActionManager = new ActionManager(actionNames, setup);
}
public MyAbstractHierarchy(MyRootDecisionSpace rds, MyModuleParams learningParams)
{
this.m_learningParams = learningParams;
m_actionLevels = new List<List<MyAction>>();
this.m_rds = rds;
for (int i = 0; i < rds.ActionManager.GetNoActions(); i++)
{
this.AddAction(i);
}
}
public MyHierarchyMaintainer(MyRootDecisionSpace rds, MyModuleParams learningParams)
: base(rds, learningParams)
{
}
public MyMotivationBasedDeleteUnselectedASM(MyModuleParams setup)
{
this.m_setup = setup;
m_rnd = new Random();
}
public VariableManager(int maxVariables, MyModuleParams setup)
{
if (maxVariables <= 0)
{
maxVariables = 1;
}
else
{
this.MAX_VARIABLES = maxVariables;
}
m_labels = new String[this.MAX_VARIABLES];
for (int i = 0; i < m_labels.Length; i++)
{
m_labels[i] = "" + i;
}
this.InitVars();
this.VarHistory = new MyVariableHistory(this, setup);
this.VarHistory.AddAllPotentialVariables();
this.ShouldBeSubspaced = new List<int>();
}
public VariableManager(String[] labels, MyModuleParams setup)
{
this.m_labels = labels;
m_vars = new MyVariable[labels.Length];
if (MAX_VARIABLES <= 0)
{
MAX_VARIABLES = 1;
}
else
{
this.MAX_VARIABLES = labels.Length;
}
this.InitVars();
this.VarHistory = new MyVariableHistory(this, setup);
this.VarHistory.AddAllPotentialVariables();
}
private List<TraceData> m_traceData; // position 0 stores the most recent state-aciton
#endregion Fields
#region Constructors
public MyEligibilityTrace(MyModuleParams learning)
{
this.m_learning = learning;
m_traceData = new List<TraceData>();
Lambda = new float[learning.TraceLength];
float tmp = learning.Lambda;
for (int i = 0; i < learning.TraceLength; i++)
{
Lambda[i] = tmp;
tmp *= learning.Lambda;
}
}
public MyLocalVariableHistory(MyRootDecisionSpace rds,
MyModuleParams setup, IDecisionSpace ds)
{
m_rds = rds;
m_setup = setup;
changes = new Dictionary<int, float>();
// all potential variables added by default
for (int i = 0; i < rds.VarManager.GetMaxVariables(); i++)
{
changes.Add(i, INIT_VAL);
}
}
public MyVariableHistory(VariableManager vm, MyModuleParams setup)
{
this.vm = vm;
this.setup = setup;
this.changed = new List<List<int>>();
isMonitored = new bool[vm.GetMaxVariables()];
}
public MyMotivationSource(MyModuleParams setup, int level)
{
this.m_setup = setup;
this.m_val = 0;
this.m_level = level;
}
public MyAbsHierarchyMaintainer(MyRootDecisionSpace rds, MyModuleParams learningParams)
: base(rds, learningParams)
{
}
