public void testFirstStepsOfQLAAgentWhenFirstStepTerminates()
{
QLearningAgent <CellWorldPosition, String> qla = new QLearningAgent <CellWorldPosition, String>(
fourByThree);
CellWorldPosition startingPosition = new CellWorldPosition(1, 4);
String action = qla.decideAction(new MDPPerception <CellWorldPosition>(
startingPosition, -0.04));
Assert.AreEqual(CellWorld.LEFT, action);
Randomizer betweenEightyANdNinetyPercent = new MockRandomizer(
new double[] { 0.85 }); // to force left to become an "up"
qla.execute(action, betweenEightyANdNinetyPercent);
Assert.AreEqual(new CellWorldPosition(2, 4), qla.getCurrentState());
Assert.AreEqual(-1.0, qla.getCurrentReward(), 0.001);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
String action2 = qla.decideAction(new MDPPerception <CellWorldPosition>(
new CellWorldPosition(2, 4), -1));
Assert.IsNull(action2);
Assert.AreEqual(-1.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
}
public void testFirstStepsOfQLAAgentUnderNormalProbability()
{
QLearningAgent <CellWorldPosition, String> qla = new QLearningAgent <CellWorldPosition, String>(
fourByThree);
Randomizer alwaysLessThanEightyPercent = new MockRandomizer(
new double[] { 0.7 });
CellWorldPosition startingPosition = new CellWorldPosition(1, 4);
String action = qla.decideAction(new MDPPerception <CellWorldPosition>(
startingPosition, -0.04));
Assert.AreEqual(CellWorld.LEFT, action);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
qla.execute(action, alwaysLessThanEightyPercent);
Assert.AreEqual(new CellWorldPosition(1, 3), qla.getCurrentState());
Assert.AreEqual(-0.04, qla.getCurrentReward(), 0.001);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
String action2 = qla.decideAction(new MDPPerception <CellWorldPosition>(
new CellWorldPosition(1, 3), -0.04));
Assert.AreEqual(-0.04, qla.getQTable().getQValue(startingPosition,
action), 0.001);
}
public void testQLearningAgent()
{
QLearningAgent <CellWorldPosition, String> qla = new QLearningAgent <CellWorldPosition, string>(
fourByThree);
Randomizer r = new MockRandomizer(new double[] { 0.1, 0.9, 0.2, 0.8,
0.3, 0.7, 0.4, 0.6, 0.5 });
// Randomizer r = new JavaRandomizer();
Dictionary <Pair <CellWorldPosition, String>, Double> q = null;
QTable <CellWorldPosition, String> qTable = null;
for (int i = 0; i < 100; i++)
{
qla.executeTrial(r);
q = qla.getQ();
qTable = qla.getQTable();
}
// qTable.normalize();
// System.Console.WriteLine(qTable);
// System.Console.WriteLine(qTable.getPolicy());
}
public void testFirstStepsOfQLAAgentWhenFirstStepTerminates()
{
QLearningAgent<CellWorldPosition, String> qla = new QLearningAgent<CellWorldPosition, String>(
fourByThree);
CellWorldPosition startingPosition = new CellWorldPosition(1, 4);
String action = qla.decideAction(new MDPPerception<CellWorldPosition>(
startingPosition, -0.04));
Assert.AreEqual(CellWorld.LEFT, action);
Randomizer betweenEightyANdNinetyPercent = new MockRandomizer(
new double[] { 0.85 }); // to force left to become an "up"
qla.execute(action, betweenEightyANdNinetyPercent);
Assert.AreEqual(new CellWorldPosition(2, 4), qla.getCurrentState());
Assert.AreEqual(-1.0, qla.getCurrentReward(), 0.001);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
String action2 = qla.decideAction(new MDPPerception<CellWorldPosition>(
new CellWorldPosition(2, 4), -1));
Assert.IsNull(action2);
Assert.AreEqual(-1.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
}
public void testFirstStepsOfQLAAgentUnderNormalProbability()
{
QLearningAgent<CellWorldPosition, String> qla = new QLearningAgent<CellWorldPosition, String>(
fourByThree);
Randomizer alwaysLessThanEightyPercent = new MockRandomizer(
new double[] { 0.7 });
CellWorldPosition startingPosition = new CellWorldPosition(1, 4);
String action = qla.decideAction(new MDPPerception<CellWorldPosition>(
startingPosition, -0.04));
Assert.AreEqual(CellWorld.LEFT, action);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
qla.execute(action, alwaysLessThanEightyPercent);
Assert.AreEqual(new CellWorldPosition(1, 3), qla.getCurrentState());
Assert.AreEqual(-0.04, qla.getCurrentReward(), 0.001);
Assert.AreEqual(0.0, qla.getQTable().getQValue(startingPosition,
action), 0.001);
String action2 = qla.decideAction(new MDPPerception<CellWorldPosition>(
new CellWorldPosition(1, 3), -0.04));
Assert.AreEqual(-0.04, qla.getQTable().getQValue(startingPosition,
action), 0.001);
}
public void testQLearningAgent()
{
QLearningAgent<CellWorldPosition, String> qla = new QLearningAgent<CellWorldPosition,string>(
fourByThree);
Randomizer r = new MockRandomizer(new double[] { 0.1, 0.9, 0.2, 0.8,
0.3, 0.7, 0.4, 0.6, 0.5 });
// Randomizer r = new JavaRandomizer();
Dictionary<Pair<CellWorldPosition, String>, Double> q = null;
QTable<CellWorldPosition, String> qTable = null;
for (int i = 0; i < 100; i++)
{
qla.executeTrial(r);
q = qla.getQ();
qTable = qla.getQTable();
}
// qTable.normalize();
// System.Console.WriteLine(qTable);
// System.Console.WriteLine(qTable.getPolicy());
}