static void valueIterationDemo()
{
System.Console.WriteLine("DEMO: Value Iteration");
System.Console.WriteLine("=====================");
System.Console.WriteLine("Figure 17.3");
System.Console.WriteLine("-----------");
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
IMarkovDecisionProcess <Cell <double>, CellWorldAction> mdp = MDPFactory.createMDPForFigure17_3(cw);
ValueIteration <Cell <double>, CellWorldAction>
vi = new ValueIteration <Cell <double>, CellWorldAction>(1.0);
IMap <Cell <double>, double> U = vi.valueIteration(mdp, 0.0001);
System.Console.WriteLine("(1,1) = " + U.Get(cw.GetCellAt(1, 1)));
System.Console.WriteLine("(1,2) = " + U.Get(cw.GetCellAt(1, 2)));
System.Console.WriteLine("(1,3) = " + U.Get(cw.GetCellAt(1, 3)));
System.Console.WriteLine("(2,1) = " + U.Get(cw.GetCellAt(2, 1)));
System.Console.WriteLine("(2,3) = " + U.Get(cw.GetCellAt(2, 3)));
System.Console.WriteLine("(3,1) = " + U.Get(cw.GetCellAt(3, 1)));
System.Console.WriteLine("(3,2) = " + U.Get(cw.GetCellAt(3, 2)));
System.Console.WriteLine("(3,3) = " + U.Get(cw.GetCellAt(3, 3)));
System.Console.WriteLine("(4,1) = " + U.Get(cw.GetCellAt(4, 1)));
System.Console.WriteLine("(4,2) = " + U.Get(cw.GetCellAt(4, 2)));
System.Console.WriteLine("(4,3) = " + U.Get(cw.GetCellAt(4, 3)));
System.Console.WriteLine("=========================");
}
public void setUp()
{
cw = CellWorldFactory.CreateCellWorldForFig17_1();
mdp = MDPFactory.createMDPForFigure17_3(cw);
pi = new PolicyIteration <Cell <double>, CellWorldAction>(
new ModifiedPolicyEvaluation <Cell <double>, CellWorldAction>(50, 1.0));
}
public void Add()
{
var addMap = new List <Cell>();
var addWorld = new CellWorld();
var r = new Random();
for (var i = 0; i < 1000; i++)
{
var x = r.Next(1000);
var y = r.Next(1000);
var n = new Cell {
X = x, Y = y
};
if (addMap.Contains(n) || _world[x, y])
{
continue;
}
addMap.Add(n);
addWorld[x, y] = true;
}
_world = _world.Add(addWorld);
foreach (var position in _testMap)
{
Assert.IsTrue(_world[position.X, position.Y]);
}
foreach (var position in addMap)
{
Assert.IsTrue(_world[position.X, position.Y]);
}
}
static void passiveTDAgentDemo()
{
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
CellWorldEnvironment cwe = new CellWorldEnvironment(
cw.GetCellAt(1, 1),
cw.GetCells(),
MDPFactory.createTransitionProbabilityFunctionForFigure17_1(cw),
CommonFactory.CreateRandom());
IMap <Cell <double>, CellWorldAction> fixedPolicy = CollectionFactory.CreateMap <Cell <double>, CellWorldAction>();
fixedPolicy.Put(cw.GetCellAt(1, 1), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(1, 2), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(1, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(2, 1), CellWorldAction.Left);
fixedPolicy.Put(cw.GetCellAt(2, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(3, 1), CellWorldAction.Left);
fixedPolicy.Put(cw.GetCellAt(3, 2), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(3, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(4, 1), CellWorldAction.Left);
PassiveTDAgent <Cell <double>, CellWorldAction> ptda
= new PassiveTDAgent <Cell <double>, CellWorldAction>(fixedPolicy, 0.2, 1.0);
cwe.AddAgent(ptda);
output_utility_learning_rates(ptda, 20, 500, 100, 1);
}
public void Remove()
{
var removeMap = new List <Cell>();
var removeWorld = new CellWorld();
var r = new Random();
for (var i = 0; i < 1000; i++)
{
var index = r.Next(_testMap.Count);
removeWorld[_testMap[index].X, _testMap[index].Y] = true;
removeMap.Add(_testMap[index]);
_testMap.RemoveAt(index);
}
_world = _world.Remove(removeWorld);
foreach (var position in _testMap)
{
Assert.IsTrue(_world[position.X, position.Y]);
}
foreach (var position in removeMap)
{
Assert.IsFalse(_world[position.X, position.Y]);
}
}
static void policyIterationDemo()
{
System.Console.WriteLine("DEMO: Policy Iteration");
System.Console.WriteLine("======================");
System.Console.WriteLine("Figure 17.3");
System.Console.WriteLine("-----------");
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
IMarkovDecisionProcess <Cell <double>, CellWorldAction> mdp = MDPFactory.createMDPForFigure17_3(cw);
PolicyIteration <Cell <double>, CellWorldAction>
pi = new PolicyIteration <Cell <double>, CellWorldAction>(
new ModifiedPolicyEvaluation <Cell <double>, CellWorldAction>(50, 1.0));
IPolicy <Cell <double>, CellWorldAction> policy = pi.policyIteration(mdp);
System.Console.WriteLine("(1,1) = " + policy.action(cw.GetCellAt(1, 1)));
System.Console.WriteLine("(1,2) = " + policy.action(cw.GetCellAt(1, 2)));
System.Console.WriteLine("(1,3) = " + policy.action(cw.GetCellAt(1, 3)));
System.Console.WriteLine("(2,1) = " + policy.action(cw.GetCellAt(2, 1)));
System.Console.WriteLine("(2,3) = " + policy.action(cw.GetCellAt(2, 3)));
System.Console.WriteLine("(3,1) = " + policy.action(cw.GetCellAt(3, 1)));
System.Console.WriteLine("(3,2) = " + policy.action(cw.GetCellAt(3, 2)));
System.Console.WriteLine("(3,3) = " + policy.action(cw.GetCellAt(3, 3)));
System.Console.WriteLine("(4,1) = " + policy.action(cw.GetCellAt(4, 1)));
System.Console.WriteLine("(4,2) = " + policy.action(cw.GetCellAt(4, 2)));
System.Console.WriteLine("(4,3) = " + policy.action(cw.GetCellAt(4, 3)));
System.Console.WriteLine("=========================");
}
/**
* Constructs an MDP that can be used to generate the utility values
* detailed in Fig 17.3.
*
* @param cw
* the cell world from figure 17.1.
* @return an MDP that can be used to generate the utility values detailed
* in Fig 17.3.
*/
public static MarkovDecisionProcess <Cell <Double>, CellWorldAction> createMDPForFigure17_3(
CellWorld <Double> cw)
{
return(new MDP <Cell <Double>, CellWorldAction>(cw.getCells(),
cw.getCellAt(1, 1), createActionsFunctionForFigure17_1(cw),
createTransitionProbabilityFunctionForFigure17_1(cw),
createRewardFunctionForFigure17_1()));
}
/// <summary>
/// Create the cell world as defined in Figure 17.1 in AIMA3e. (a) A simple 4
/// x 3 environment that presents the agent with a sequential decision
/// problem.
/// </summary>
/// <returns>a cell world representation of Fig 17.1 in AIMA3e.</returns>
public static CellWorld <double> CreateCellWorldForFig17_1()
{
CellWorld <double> cw = new CellWorld <double>(4, 3, -0.04);
cw.RemoveCell(2, 2);
cw.GetCellAt(4, 3).setContent(1.0);
cw.GetCellAt(4, 2).setContent(-1.0);
return(cw);
}
/**
* Create the cell world as defined in Figure 17.1 in AIMA3e. (a) A simple 4
* x 3 environment that presents the agent with a sequential decision
* problem.
*
* @return a cell world representation of Fig 17.1 in AIMA3e.
*/
public static CellWorld<Double> createCellWorldForFig17_1()
{
CellWorld<Double> cw = new CellWorld<Double>(4, 3, -0.04);
cw.removeCell(2, 2);
cw.getCellAt(4, 3).setContent(1.0);
cw.getCellAt(4, 2).setContent(-1.0);
return cw;
}
createTransitionProbabilityFunctionForFigure17_1(
CellWorld <Double> cw)
{
TransitionProbabilityFunction <Cell <Double>, CellWorldAction> tf =
new TransitionProbabilityFunctionImpl(cw);
{
}
;
return(tf);
}
public static MDP <CellWorldPosition, String> CreateFourByThreeMDP()
{
CellWorld cw = new CellWorld(3, 4, 0.4);
cw = new CellWorld(3, 4, -0.04);
cw.MarkBlocked(2, 2);
cw.SetTerminalState(2, 4);
cw.SetReward(2, 4, -1);
cw.SetTerminalState(3, 4);
cw.SetReward(3, 4, 1);
return(cw.AsMdp());
}
public static MDP<CellWorldPosition, String> createFourByThreeMDP()
{
CellWorld cw = new CellWorld(3, 4, 0.4);
cw = new CellWorld(3, 4, -0.04);
cw.markBlocked(2, 2);
cw.setTerminalState(2, 4);
cw.setReward(2, 4, -1);
cw.setTerminalState(3, 4);
cw.setReward(3, 4, 1);
return cw.asMdp();
}
public void setUp() {
cw = new CellWorld(3, 4, -0.04);
cw.markBlocked(2, 2);
cw.setTerminalState(2, 4);
cw.setReward(2, 4, -1);
cw.setTerminalState(3, 4);
cw.setReward(3, 4, 1);
alwaysLessThanEightyPercent = new MockRandomizer(new double[] { 0.7 });
betweenEightyAndNinetyPercent = new MockRandomizer(
new double[] { 0.85 });
greaterThanNinetyPercent = new MockRandomizer(new double[] { 0.95 });
}
/**
* Returns the allowed actions from a specified cell within the cell world
* described in Fig 17.1.
*
* @param cw
* the cell world from figure 17.1.
* @return the set of actions allowed at a particular cell. This set will be
* empty if at a terminal state.
*/
public static ActionsFunction <Cell <Double>, CellWorldAction> createActionsFunctionForFigure17_1(
CellWorld <Double> cw)
{
Set <Cell <Double> > terminals = new Set <Cell <Double> >();
terminals.add(cw.getCellAt(4, 3));
terminals.add(cw.getCellAt(4, 2));
ActionsFunction <Cell <Double>, CellWorldAction> af = null;
//new ActionsFunction<Cell<Double>, CellWorldAction>() {
//};
// TODO
return(af);
}
static void qLearningAgentDemo()
{
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
CellWorldEnvironment cwe = new CellWorldEnvironment(
cw.GetCellAt(1, 1),
cw.GetCells(),
MDPFactory.createTransitionProbabilityFunctionForFigure17_1(cw),
CommonFactory.CreateRandom());
QLearningAgent <Cell <double>, CellWorldAction> qla = new QLearningAgent <Cell <double>, CellWorldAction>(
MDPFactory.createActionsFunctionForFigure17_1(cw),
CellWorldAction.None, 0.2, 1.0, 5,
2.0);
cwe.AddAgent(qla);
output_utility_learning_rates(qla, 20, 10000, 500, 20);
}
public void Prepare()
{
_testMap = new List <Cell>
{
};
for (var i = 0; i < 100; i++)
{
for (var j = 0; j < 100; j++)
{
_testMap.Add(new Cell {
X = i, Y = j
});
}
}
var r = new Random();
for (var i = 0; i < 10000; i++)
{
var x = r.Next(1000);
var y = r.Next(1000);
var n = new Cell {
X = 100 + x, Y = 100 + y
};
if (_testMap.Contains(n))
{
continue;
}
_testMap.Add(n);
}
_world = new CellWorld();
foreach (var position in _testMap)
{
_world[position.X, position.Y] = true;
}
}
static void passiveADPAgentDemo()
{
System.Console.WriteLine("=======================");
System.Console.WriteLine("DEMO: Passive-ADP-Agent");
System.Console.WriteLine("=======================");
System.Console.WriteLine("Figure 21.3");
System.Console.WriteLine("-----------");
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
CellWorldEnvironment cwe = new CellWorldEnvironment(
cw.GetCellAt(1, 1),
cw.GetCells(),
MDPFactory.createTransitionProbabilityFunctionForFigure17_1(cw),
CommonFactory.CreateRandom());
IMap <Cell <double>, CellWorldAction> fixedPolicy = CollectionFactory.CreateInsertionOrderedMap <Cell <double>, CellWorldAction>();
fixedPolicy.Put(cw.GetCellAt(1, 1), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(1, 2), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(1, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(2, 1), CellWorldAction.Left);
fixedPolicy.Put(cw.GetCellAt(2, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(3, 1), CellWorldAction.Left);
fixedPolicy.Put(cw.GetCellAt(3, 2), CellWorldAction.Up);
fixedPolicy.Put(cw.GetCellAt(3, 3), CellWorldAction.Right);
fixedPolicy.Put(cw.GetCellAt(4, 1), CellWorldAction.Left);
PassiveADPAgent <Cell <double>, CellWorldAction> padpa = new PassiveADPAgent <Cell <double>, CellWorldAction>(
fixedPolicy, cw.GetCells(), cw.GetCellAt(1, 1),
MDPFactory.createActionsFunctionForFigure17_1(cw),
new ModifiedPolicyEvaluation <Cell <double>, CellWorldAction>(10, 1.0));
cwe.AddAgent(padpa);
output_utility_learning_rates(padpa, 20, 100, 100, 1);
System.Console.WriteLine("=========================");
}
public void setUp()
{
cw = CellWorldFactory.CreateCellWorldForFig17_1();
mdp = MDPFactory.createMDPForFigure17_3(cw);
vi = new ValueIteration<Cell<double>, CellWorldAction>(1.0);
}
/**
* Figure 17.1 (b) Illustration of the transition model of the environment:
* the 'intended' outcome occurs with probability 0.8, but with probability
* 0.2 the agent moves at right angles to the intended direction. A
* collision with a wall results in no movement.
*
* @param cw
* the cell world from figure 17.1.
* @return the transition probability function as described in figure 17.1.
*/
public static ITransitionProbabilityFunction <Cell <double>, CellWorldAction> createTransitionProbabilityFunctionForFigure17_1(CellWorld <double> cw)
{
ITransitionProbabilityFunction <Cell <double>, CellWorldAction> tf = new createTransitionProbabilityFunctionForFigure17_1TransitionProbabilityFunction(cw);
return(tf);
}
/**
* Returns the allowed actions from a specified cell within the cell world
* described in Fig 17.1.
*
* @param cw
* the cell world from figure 17.1.
* @return the set of actions allowed at a particular cell. This set will be
* empty if at a terminal state.
*/
public static IActionsFunction <Cell <double>, CellWorldAction> createActionsFunctionForFigure17_1(CellWorld <double> cw)
{
ISet <Cell <double> > terminals = CollectionFactory.CreateSet <Cell <double> >();
terminals.Add(cw.GetCellAt(4, 3));
terminals.Add(cw.GetCellAt(4, 2));
IActionsFunction <Cell <double>, CellWorldAction> af = new createActionsFunctionForFigure17_1ActionsFunction(terminals);
return(af);
}
public void setUp()
{
cw = CellWorldFactory.CreateCellWorldForFig17_1();
}
public createTransitionProbabilityFunctionForFigure17_1TransitionProbabilityFunction(CellWorld <double> cw)
{
this.cw = cw;
}
//
// PRIVATE METHODS
//
private void assertPolicyReccomends(CellWorld cw,
MDPUtilityFunction<CellWorldPosition> uf, int x, int y,
String actionExpected) {
Pair<String, Double> p = cw.getTransitionModel()
.getTransitionWithMaximumExpectedUtility(
new CellWorldPosition(x, y), uf);
Assert.assertEquals(actionExpected, p.getFirst());
}
public Example_AForge_Animat(Base parent)
: base(parent)
{
// GUI
{
m_TopPanel = new Base(this);
{
m_TopPanel.Dock = Pos.Top;
m_TopPanel.Height = 30;
Alt.GUI.Temporary.Gwen.Control.Label label = new Alt.GUI.Temporary.Gwen.Control.Label(m_TopPanel);
label.Location = new PointI(0, 2);
label.AutoSizeToContents = true;
label.TextColor = Color.LightGreen;
label.Text = "Map";
label = new Alt.GUI.Temporary.Gwen.Control.Label(m_TopPanel);
label.Location = new PointI(33, 2);
label.Text = "World size:";
label.AutoSizeToContents = true;
worldSizeBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(m_TopPanel);
worldSizeBox.Location = new PointI(102, 0);
worldSizeBox.ReadOnly = true;
worldSizeBox.Width = 40;
loadButton1 = new Alt.GUI.Temporary.Gwen.Control.Button(m_TopPanel);
loadButton1.Location = new PointI(160, 0);
loadButton1.Width = 80;
loadButton1.Text = "Sample 1";
loadButton1.Click += new System.EventHandler(loadButton1_Click);
loadButton1.NormalTextColor = Color.Red * 0.8;
loadButton2 = new Alt.GUI.Temporary.Gwen.Control.Button(m_TopPanel);
loadButton2.Location = new PointI(250, 0);
loadButton2.Width = 80;
loadButton2.Text = "Sample 2";
loadButton2.Click += new System.EventHandler(loadButton2_Click);
loadButton2.NormalTextColor = Color.Brown;
loadButton3 = new Alt.GUI.Temporary.Gwen.Control.Button(m_TopPanel);
loadButton3.Location = new PointI(340, 0);
loadButton3.Width = 80;
loadButton3.Text = "Test 1";
loadButton3.Click += new System.EventHandler(loadButton3_Click);
loadButton3.NormalTextColor = Color.Green;
loadButton4 = new Alt.GUI.Temporary.Gwen.Control.Button(m_TopPanel);
loadButton4.Location = new PointI(430, 0);
loadButton4.Width = 80;
loadButton4.Text = "Test 2";
loadButton4.Click += new System.EventHandler(loadButton4_Click);
loadButton4.NormalTextColor = Color.Blue;
loadButton5 = new Alt.GUI.Temporary.Gwen.Control.Button(m_TopPanel);
loadButton5.Location = new PointI(520, 0);
loadButton5.Width = 80;
loadButton5.Text = "Test 3";
loadButton5.Click += new System.EventHandler(loadButton5_Click);
loadButton5.NormalTextColor = Color.Violet * 0.8;
}
m_RightPanel = new Base(this);
{
m_RightPanel.Dock = Pos.Right;
m_RightPanel.Width = 140;
m_RightPanel.Margin = new Margin(5, 0, 0, 0);
Alt.GUI.Temporary.Gwen.Control.GroupBox groupBox = new Alt.GUI.Temporary.Gwen.Control.GroupBox(m_RightPanel);
{
groupBox.Location = new PointI(0, 0);
groupBox.Size = new SizeI(140, 363);
groupBox.Text = "Settings";
Alt.GUI.Temporary.Gwen.Control.Label label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 10);
label.AutoSizeToContents = true;
label.Text = "Learning algorithm:";
algorithmCombo = new Alt.GUI.Temporary.Gwen.Control.ComboBox(groupBox);
algorithmCombo.Location = new PointI(0, 30);
algorithmCombo.Width = 130;
algorithmCombo.AddItem("Q-Learning").UserData = 0;
algorithmCombo.AddItem("Sarsa").UserData = 1;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 65);
label.AutoSizeToContents = true;
label.Text = "Initial exploration rate:";
explorationRateBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
explorationRateBox.Location = new PointI(0, 85);
explorationRateBox.Width = 130;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 110);
label.AutoSizeToContents = true;
label.Text = "Initial learning rate:";
learningRateBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
learningRateBox.Location = new PointI(0, 130);
learningRateBox.Width = 130;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 155);
label.AutoSizeToContents = true;
label.Text = "Learning iterations:";
iterationsBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
iterationsBox.Location = new PointI(0, 175);
iterationsBox.Width = 130;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 210);
label.AutoSizeToContents = true;
label.Text = "Move reward:";
moveRewardBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
moveRewardBox.Location = new PointI(0, 230);
moveRewardBox.Width = 130;
moveRewardBox.ReadOnly = true;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 255);
label.AutoSizeToContents = true;
label.Text = "Wall reward:";
wallRewardBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
wallRewardBox.Location = new PointI(0, 275);
wallRewardBox.Width = 130;
wallRewardBox.ReadOnly = true;
label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 300);
label.AutoSizeToContents = true;
label.Text = "Goal reward:";
goalRewardBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
goalRewardBox.Location = new PointI(0, 320);
goalRewardBox.Width = 130;
goalRewardBox.ReadOnly = true;
}
groupBox = new Alt.GUI.Temporary.Gwen.Control.GroupBox(m_RightPanel);
{
groupBox.Location = new PointI(0, 380);
groupBox.Size = new SizeI(140, 173);
groupBox.Text = "Learning";
Alt.GUI.Temporary.Gwen.Control.Label label = new Alt.GUI.Temporary.Gwen.Control.Label(groupBox);
label.Location = new PointI(0, 10);
label.AutoSizeToContents = true;
label.Text = "Iteration:";
iterationBox = new Alt.GUI.Temporary.Gwen.Control.TextBox(groupBox);
iterationBox.Location = new PointI(0, 30);
iterationBox.Width = 130;
iterationBox.ReadOnly = true;
startLearningButton = new Alt.GUI.Temporary.Gwen.Control.Button(groupBox);
startLearningButton.Location = new PointI(5, 70);
startLearningButton.Width = 120;
startLearningButton.Text = "Start";
startLearningButton.Click += new System.EventHandler(startLearningButton_Click);
startLearningButton.NormalTextColor = Color.Green;
stopButton = new Alt.GUI.Temporary.Gwen.Control.Button(groupBox);
stopButton.Location = new PointI(5, 100);
stopButton.Width = 120;
stopButton.Text = "Stop";
stopButton.Click += new System.EventHandler(stopButton_Click);
stopButton.NormalTextColor = Color.Red;
showSolutionButton = new Alt.GUI.Temporary.Gwen.Control.Button(groupBox);
showSolutionButton.Location = new PointI(5, 130);
showSolutionButton.Width = 120;
showSolutionButton.Text = "Show solution";
showSolutionButton.Click += new System.EventHandler(showSolutionButton_Click);
showSolutionButton.NormalTextColor = Color.Blue;
}
}
cellWorld = new CellWorld(this);
cellWorld.Coloring = null;
cellWorld.Map = null;
}
}
protected static void output_utility_learning_rates(
ReinforcementAgent <Cell <double>, CellWorldAction> reinforcementAgent,
int numRuns, int numTrialsPerRun, int rmseTrialsToReport,
int reportEveryN)
{
if (rmseTrialsToReport > (numTrialsPerRun / reportEveryN))
{
throw new IllegalArgumentException("Requesting to report too many RMSE trials, max allowed for args is "
+ (numTrialsPerRun / reportEveryN));
}
CellWorld <double> cw = CellWorldFactory.CreateCellWorldForFig17_1();
CellWorldEnvironment cwe = new CellWorldEnvironment(
cw.GetCellAt(1, 1),
cw.GetCells(),
MDPFactory.createTransitionProbabilityFunctionForFigure17_1(cw),
CommonFactory.CreateRandom());
cwe.AddAgent(reinforcementAgent);
IMap <int, ICollection <IMap <Cell <double>, double> > > runs = CollectionFactory.CreateInsertionOrderedMap <int, ICollection <IMap <Cell <double>, double> > >();
for (int r = 0; r < numRuns; r++)
{
reinforcementAgent.reset();
ICollection <IMap <Cell <double>, double> > trials = CollectionFactory.CreateQueue <IMap <Cell <double>, double> >();
for (int t = 0; t < numTrialsPerRun; t++)
{
cwe.executeTrial();
if (0 == t % reportEveryN)
{
IMap <Cell <double>, double> u = reinforcementAgent
.getUtility();
//if (null == u.Get(cw.getCellAt(1, 1)))
//{
// throw new IllegalStateException(
// "Bad Utility State Encountered: r=" + r
// + ", t=" + t + ", u=" + u);
//}
trials.Add(u);
}
}
runs.Put(r, trials);
}
IStringBuilder v4_3 = TextFactory.CreateStringBuilder();
IStringBuilder v3_3 = TextFactory.CreateStringBuilder();
IStringBuilder v1_3 = TextFactory.CreateStringBuilder();
IStringBuilder v1_1 = TextFactory.CreateStringBuilder();
IStringBuilder v3_2 = TextFactory.CreateStringBuilder();
IStringBuilder v2_1 = TextFactory.CreateStringBuilder();
for (int t = 0; t < (numTrialsPerRun / reportEveryN); t++)
{
// Use the last run
IMap <Cell <double>, double> u = runs.Get(numRuns - 1).Get(t);
v4_3.Append((u.ContainsKey(cw.GetCellAt(4, 3)) ? u.Get(cw
.GetCellAt(4, 3)) : 0.0) + "\t");
v3_3.Append((u.ContainsKey(cw.GetCellAt(3, 3)) ? u.Get(cw
.GetCellAt(3, 3)) : 0.0) + "\t");
v1_3.Append((u.ContainsKey(cw.GetCellAt(1, 3)) ? u.Get(cw
.GetCellAt(1, 3)) : 0.0) + "\t");
v1_1.Append((u.ContainsKey(cw.GetCellAt(1, 1)) ? u.Get(cw
.GetCellAt(1, 1)) : 0.0) + "\t");
v3_2.Append((u.ContainsKey(cw.GetCellAt(3, 2)) ? u.Get(cw
.GetCellAt(3, 2)) : 0.0) + "\t");
v2_1.Append((u.ContainsKey(cw.GetCellAt(2, 1)) ? u.Get(cw
.GetCellAt(2, 1)) : 0.0) + "\t");
}
IStringBuilder rmseValues = TextFactory.CreateStringBuilder();
for (int t = 0; t < rmseTrialsToReport; t++)
{
// Calculate the Root Mean Square Error for utility of 1,1
// for this trial# across all runs
double xSsquared = 0;
for (int r = 0; r < numRuns; r++)
{
IMap <Cell <double>, double> u = runs.Get(r).Get(t);
double val1_1 = u.Get(cw.GetCellAt(1, 1));
//if (null == val1_1)
//{
// throw new IllegalStateException(
// "U(1,1,) is not present: r=" + r + ", t=" + t
// + ", runs.size=" + runs.Size()
// + ", runs(r).Size()=" + runs.Get(r).Size()
// + ", u=" + u);
//}
xSsquared += System.Math.Pow(0.705 - val1_1, 2);
}
double rmse = System.Math.Sqrt(xSsquared / runs.Size());
rmseValues.Append(rmse);
rmseValues.Append("\t");
}
System.Console
.WriteLine("Note: You may copy and paste the following lines into a spreadsheet to generate graphs of learning rate and RMS error in utility:");
System.Console.WriteLine("(4,3)" + "\t" + v4_3);
System.Console.WriteLine("(3,3)" + "\t" + v3_3);
System.Console.WriteLine("(1,3)" + "\t" + v1_3);
System.Console.WriteLine("(1,1)" + "\t" + v1_1);
System.Console.WriteLine("(3,2)" + "\t" + v3_2);
System.Console.WriteLine("(2,1)" + "\t" + v2_1);
System.Console.WriteLine("RMSeiu" + "\t" + rmseValues);
}
public void setUp()
{
cw = CellWorldFactory.CreateCellWorldForFig17_1();
mdp = MDPFactory.createMDPForFigure17_3(cw);
}
public TransitionProbabilityFunctionImpl(CellWorld <Double> cw)
{
this.cw = cw;
}
