public Clarion_AGC_Model(Agent MAKaey) { World.LoggingLevel = System.Diagnostics.TraceLevel.Off; this.MAKaey = MAKaey; NeuralNetwork = AgentInitializer.InitializeImplicitDecisionNetwork(MAKaey, SimplifiedQBPNetwork.Factory); }
public static void Initialize() { World.LoggingLevel = TraceLevel.Off; John = World.NewAgent("John"); SimplifiedQBPNetwork net = AgentInitializer.InitializeImplicitDecisionNetwork(John, SimplifiedQBPNetwork.Factory); net.Input.Add(World.NewDimensionValuePair("Boolean 1", true)); net.Input.Add(World.NewDimensionValuePair("Boolean 1", false)); net.Input.Add(World.NewDimensionValuePair("Boolean 2", true)); net.Input.Add(World.NewDimensionValuePair("Boolean 2", false)); net.Output.Add(World.NewExternalActionChunk(true)); net.Output.Add(World.NewExternalActionChunk(false)); John.ACS.Parameters.PERFORM_RER_REFINEMENT = false; John.ACS.Parameters.SELECTION_TEMPERATURE = .01; John.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; John.ACS.Parameters.FIXED_IRL_LEVEL_SELECTION_MEASURE = 0; John.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 0; //Tweak these parameters to see the impact each has on accuracy and learning net.Parameters.LEARNING_RATE = 1; net.Parameters.MOMENTUM = .02; John.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = .5; John.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = .5; John.Commit(net); }
private void Initialize() { World.LoggingLevel = TraceLevel.Off; p1 = World.NewDimensionValuePair("Peg", 1); p2 = World.NewDimensionValuePair("Peg", 2); p3 = World.NewDimensionValuePair("Peg", 3); p4 = World.NewDimensionValuePair("Peg", 4); p5 = World.NewDimensionValuePair("Peg", 5); mp1 = World.NewExternalActionChunk(); mp2 = World.NewExternalActionChunk(); mp3 = World.NewExternalActionChunk(); mp4 = World.NewExternalActionChunk(); mp5 = World.NewExternalActionChunk(); mp1 += p1; mp2 += p2; mp3 += p3; mp4 += p4; mp5 += p5; John = World.NewAgent(); net = AgentInitializer.InitializeImplicitDecisionNetwork(John, SimplifiedQBPNetwork.Factory); net.Input.Add(p1); net.Input.Add(p2); net.Input.Add(p3); net.Input.Add(p4); net.Input.Add(p5); net.Output.Add(mp1); net.Output.Add(mp2); net.Output.Add(mp3); net.Output.Add(mp4); net.Output.Add(mp5); net.Parameters.LEARNING_RATE = 1; net.Parameters.MOMENTUM = .01; John.Commit(net); RefineableActionRule.GlobalParameters.GENERALIZATION_THRESHOLD_1 = -.01; RefineableActionRule.GlobalParameters.SPECIALIZATION_THRESHOLD_1 = -.4; }
static void Main(string[] args) { //Initialize the task Console.WriteLine("Initializing the Simple Hello World Task"); int CorrectCounter = 0; int NumberTrials = 10000; int progress = 0; World.LoggingLevel = TraceLevel.Off; TextWriter orig = Console.Out; StreamWriter sw = File.CreateText("HelloWorldSimple.txt"); DimensionValuePair hi = World.NewDimensionValuePair("Salutation", "Hellow"); DimensionValuePair bye = World.NewDimensionValuePair("Salutation", "Goodbyew"); ExternalActionChunk sayHi = World.NewExternalActionChunk("Hello"); ExternalActionChunk sayBye = World.NewExternalActionChunk("Goodbye"); //Initialize the Agent Agent John = World.NewAgent("John"); SimplifiedQBPNetwork net = AgentInitializer.InitializeImplicitDecisionNetwork(John, SimplifiedQBPNetwork.Factory); net.Input.Add(hi); net.Input.Add(bye); net.Output.Add(sayHi); net.Output.Add(sayBye); John.Commit(net); net.Parameters.LEARNING_RATE = 1; John.ACS.Parameters.PERFORM_RER_REFINEMENT = false; //Run the task Console.WriteLine("Running the Simple Hello World Task"); Console.SetOut(sw); Random rand = new Random(); SensoryInformation si; ExternalActionChunk chosen; for (int i = 0; i < NumberTrials; i++) { si = World.NewSensoryInformation(John); //Randomly choose an input to perceive. if (rand.NextDouble() < .5) { //Say "Hello" si.Add(hi, John.Parameters.MAX_ACTIVATION); si.Add(bye, John.Parameters.MIN_ACTIVATION); } else { //Say "Goodbye" si.Add(hi, John.Parameters.MIN_ACTIVATION); si.Add(bye, John.Parameters.MAX_ACTIVATION); } //Perceive the sensory information John.Perceive(si); //Choose an action chosen = John.GetChosenExternalAction(si); //Deliver appropriate feedback to the agent if (chosen == sayHi) { //The agent said "Hello". if (si[hi] == John.Parameters.MAX_ACTIVATION) { //The agent responded correctly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was correct"); //Record the agent's success. CorrectCounter++; //Give positive feedback. John.ReceiveFeedback(si, 1.0); } else { //The agent responded incorrectly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was incorrect"); //Give negative feedback. John.ReceiveFeedback(si, 0.0); } } else { //The agent said "Goodbye". if (si[bye] == John.Parameters.MAX_ACTIVATION) { //The agent responded correctly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was correct"); //Record the agent's success. CorrectCounter++; //Give positive feedback. John.ReceiveFeedback(si, 1.0); } else { //The agent responded incorrectly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was incorrect"); //Give negative feedback. John.ReceiveFeedback(si, 0.0); } } Console.SetOut(orig); progress = (int)(((double)(i + 1) / (double)NumberTrials) * 100); Console.CursorLeft = 0; Console.Write(progress + "% Complete.."); Console.SetOut(sw); } //Report Results Console.WriteLine("Reporting Results for the Simple Hello World Task"); Console.WriteLine("John got " + CorrectCounter + " correct out of " + NumberTrials + " trials (" + (int)Math.Round(((double)CorrectCounter / (double)NumberTrials) * 100) + "%)"); Console.WriteLine("At the end of the task, John had learned the following rules:"); foreach (var i in John.GetInternals(Agent.InternalContainers.ACTION_RULES)) { Console.WriteLine(i); } sw.Close(); Console.SetOut(orig); Console.CursorLeft = 0; Console.WriteLine("100% Complete.."); //Kill the agent to end the task Console.WriteLine("Killing John to end the program"); John.Die(); Console.WriteLine("John is Dead"); Console.WriteLine("The Simple Hello World Task has finished"); Console.WriteLine("The results have been saved to \"HelloWorldSimple.txt\""); Console.Write("Press any key to exit"); Console.ReadKey(true); }
public void Initialize() { // Dimension Value Pairs: sayWhat = World.NewDimensionValuePair("YourAction", "What do you want to do?"); // External Action Chunks: sayCooperate = World.NewExternalActionChunk("Cooperate"); sayDefect = World.NewExternalActionChunk("Defect"); // placeholder // GoalChunk salute = World.NewGoalChunk("Salute"); // GoalChunk bidFarewell = World.NewGoalChunk("Bid Farewell"); // WM Actions: wmuacC = World.NewWorkingMemoryUpdateActionChunk("Remember my opponent cooperated"); wmuacD = World.NewWorkingMemoryUpdateActionChunk("Remember my opponent defected"); DeclarativeChunk dcoc = World.NewDeclarativeChunk("My opponent cooperated"); DeclarativeChunk dcod = World.NewDeclarativeChunk("My opponent defected"); wmuacC.Add(WorkingMemory.RecognizedActions.SET_RESET, dcoc); wmuacD.Add(WorkingMemory.RecognizedActions.SET_RESET, dcod); // Set up a two agent model (meaning two agents with the same setup, playing against each other) Alice = World.NewAgent("Alice"); Bob = World.NewAgent("Bob"); // Simulating environment will determine inputs to each agent based on what each agent does.. // Feedback is determined by payoff matrix.. payoff = new int [2, 2, 2]; // Doing this the hard way. Could set this up all in-line above, but this makes the table // more explicit in terms of how we want to use it. // The payoff matrix here is called "Friend or Foe", about the simplest case // indices mean: FOR-WHICH-AGENT, WHAT-ALICE-DOES, WHAT-BOB-DOES payoff[_ALICE, _COOPERATE, _COOPERATE] = 1; payoff[_ALICE, _COOPERATE, _DEFECT] = 0; payoff[_ALICE, _DEFECT, _COOPERATE] = 2; payoff[_ALICE, _DEFECT, _DEFECT] = 0; payoff[_BOB, _COOPERATE, _COOPERATE] = 1; payoff[_BOB, _COOPERATE, _DEFECT] = 2; payoff[_BOB, _DEFECT, _COOPERATE] = 0; payoff[_BOB, _DEFECT, _DEFECT] = 0; maxpay = 2; results = new int[_TRIALS, 2, 2]; // Set up a Q-learning Net = // -- Eligibility Condition = True if "What do you want to do?" is in input, otherwise False // -- Input = "My opponent cooperated", "My opponent defected", "What do you want to do?" // -- Output = "I want to defect", "I want to cooperate" // // Also, RER is turned ON QBPNetwork net_A = AgentInitializer.InitializeImplicitDecisionNetwork(Alice, QBPNetwork.Factory, QNetEC); net_A.Input.Add(sayWhat); net_A.Input.Add(sayCooperate); net_A.Input.Add(sayDefect); net_A.Output.Add(sayCooperate); net_A.Output.Add(sayDefect); Alice.Commit(net_A); net_A.Parameters.LEARNING_RATE = 1; Alice.ACS.Parameters.PERFORM_RER_REFINEMENT = true; // it's true by default anyway Alice.ACS.Parameters.LEVEL_SELECTION_METHOD = ActionCenteredSubsystem.LevelSelectionMethods.COMBINED; Alice.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; Alice.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 1; Alice.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 1; Alice.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 1; Alice.ACS.Parameters.WM_UPDATE_ACTION_PROBABILITY = 1; // Rules (2 rules) = // Rule 1: // -- Condition = "Your opponent cooperated" // -- Action = Set "My opponent cooperated" in WM // Rule 2: // -- Condition = "Your opponent defected" // -- Action = Set "My opponent defect" in WM FixedRule ruleA1 = AgentInitializer.InitializeActionRule(Alice, FixedRule.Factory, wmuacC, FRSC); FixedRule ruleA2 = AgentInitializer.InitializeActionRule(Alice, FixedRule.Factory, wmuacD, FRSC); Alice.Commit(ruleA1); Alice.Commit(ruleA2); QBPNetwork net_B = AgentInitializer.InitializeImplicitDecisionNetwork(Bob, QBPNetwork.Factory, QNetEC); net_B.Input.Add(sayWhat); net_B.Input.Add(sayCooperate); net_B.Input.Add(sayDefect); net_B.Output.Add(sayCooperate); net_B.Output.Add(sayDefect); Bob.Commit(net_B); // Use Weighted Combination // NO partial match on TL net_B.Parameters.LEARNING_RATE = 1; Bob.ACS.Parameters.PERFORM_RER_REFINEMENT = true; Bob.ACS.Parameters.LEVEL_SELECTION_METHOD = ActionCenteredSubsystem.LevelSelectionMethods.COMBINED; Bob.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; Bob.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 1; Bob.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 1; Bob.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 1; Bob.ACS.Parameters.WM_UPDATE_ACTION_PROBABILITY = 1; FixedRule ruleB1 = AgentInitializer.InitializeActionRule(Bob, FixedRule.Factory, wmuacC, FRSC); FixedRule ruleB2 = AgentInitializer.InitializeActionRule(Bob, FixedRule.Factory, wmuacD, FRSC); Bob.Commit(ruleB1); Bob.Commit(ruleB2); // Initially using the same parameters for RER as Full Hello World RefineableActionRule.GlobalParameters.SPECIALIZATION_THRESHOLD_1 = -.6; RefineableActionRule.GlobalParameters.GENERALIZATION_THRESHOLD_1 = -.1; RefineableActionRule.GlobalParameters.INFORMATION_GAIN_OPTION = RefineableActionRule.IGOptions.PERFECT; /* * Note -- What should be seems is that when you pass in "Your opponent…", * the agent should return the "Do Nothing" external action * (since it performed an internal WM action).. * However, you can just ignore this either way.. */ }
public void Init() { MyAgent = World.NewAgent(MyHero.Name); // set up the network and learning parameters for the agent SimplifiedQBPNetwork net = AgentInitializer.InitializeImplicitDecisionNetwork(MyAgent, SimplifiedQBPNetwork.Factory); net = AgentInitializer.InitializeImplicitDecisionNetwork(MyAgent, SimplifiedQBPNetwork.Factory); for (int i = 0; i < HealthInputs.GetLength(0); i++) { // net.Input.Add(HealthInputs[i]); // net.Input.Add(ManaInputs[i]); } for (int i = 0; i < InventoryInputs.GetLength(0); i++) { for (int j = 0; j < InventoryInputs.GetLength(1); j++) { net.Input.Add(InventoryInputs[i, j]); } net.Output.Add(PurchaseActions[i]); } FixedRule fr; for (int i = 0; i < PurchaseActions.Length; i++) { fr = AgentInitializer.InitializeActionRule(MyAgent, FixedRule.Factory, PurchaseActions[i], ImitativeSupportDelegate); // fr.GeneralizedCondition.Add(Inputs[i], true); MyAgent.Commit(fr); } // AddRules(); //net.Parameters.LEARNING_RATE = 0.5; net.Parameters.LEARNING_RATE = 2.0; // net.Parameters.MOMENTUM = 0.02; MyAgent.Commit(net); MyAgent.ACS.Parameters.SELECTION_TEMPERATURE = 0.05; MyAgent.ACS.Parameters.DELETION_FREQUENCY = 100; MyAgent.ACS.Parameters.LEVEL_SELECTION_METHOD = ActionCenteredSubsystem.LevelSelectionMethods.COMBINED; MyAgent.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; // set up the probabilities used to select which system will be chosen to select an action // at each step (should total 1.0): // BL - bottom layer (reinforcement learning neural net) // RER - rule extraction and refinement - extracts rules from the bottom layer // IRL - independent rule learning - does not use the bottom layer for learning rules // FR - fixed rules - Clarion cannot change these (though they can be added/removed externally) // We are currently using fixed rules when we want the agent to immitate the human player and train // the bottom layer MyAgent.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 0.33; MyAgent.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 0.33; MyAgent.ACS.Parameters.FIXED_IRL_LEVEL_SELECTION_MEASURE = 0; MyAgent.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 0.33; /* * MyAgent.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 0.75; * MyAgent.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 0.25; * MyAgent.ACS.Parameters.FIXED_IRL_LEVEL_SELECTION_MEASURE = 0; * MyAgent.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 0; */ //MyAgent.ACS.Parameters.VARIABLE_BL_BETA = 0.5; //MyAgent.ACS.Parameters.VARIABLE_RER_BETA = 0.5; //MyAgent.ACS.Parameters.VARIABLE_IRL_BETA = 0; //MyAgent.ACS.Parameters.VARIABLE_FR_BETA = 0; }
public static void InitializeAgent(Groups gr) { Participant = World.NewAgent(); BPNetwork idn = AgentInitializer.InitializeImplicitDecisionNetwork(Participant, BPNetwork.Factory); idn.Input.AddRange(dvs); idn.Output.AddRange(acts); Participant.Commit(idn); foreach (DeclarativeChunk t in tools) { RefineableActionRule a = AgentInitializer.InitializeActionRule(Participant, RefineableActionRule.Factory, World.GetActionChunk("Tool")); foreach (DimensionValuePair dv in t) { a.GeneralizedCondition.Add(dv, true); } Participant.Commit(a); } foreach (DeclarativeChunk g in guns) { RefineableActionRule a = AgentInitializer.InitializeActionRule(Participant, RefineableActionRule.Factory, World.GetActionChunk("Gun")); foreach (DimensionValuePair dv in g) { a.GeneralizedCondition.Add(dv, true); } Participant.Commit(a); } Participant.ACS.Parameters.PERFORM_RER_REFINEMENT = false; Participant.ACS.Parameters.PERFORM_DELETION_BY_DENSITY = false; Participant.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 1; Participant.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 1; Participant.ACS.Parameters.FIXED_IRL_LEVEL_SELECTION_MEASURE = 0; Participant.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 0; Participant.ACS.Parameters.B = 1; HonorDrive honor = AgentInitializer.InitializeDrive(Participant, HonorDrive.Factory, r.NextDouble()); GenericEquation hd = AgentInitializer.InitializeDriveComponent(honor, GenericEquation.Factory, (Equation)TangentEquation); var ins = Drive.GenerateTypicalInputs(honor); ParameterChangeActionChunk pac = World.NewParameterChangeActionChunk(); pac.Add(Participant.ACS, "MCS_RER_SELECTION_MEASURE", .5); hd.Input.AddRange(ins); hd.Parameters.MAX_ACTIVATION = 5; honor.Commit(hd); honor.Parameters.DRIVE_GAIN = (gr == Groups.PRIVATE) ? .1 / 5 : .2 / 5; Participant.Commit(honor); ParameterSettingModule lpm = AgentInitializer.InitializeMetaCognitiveModule(Participant, ParameterSettingModule.Factory); ACSLevelProbabilitySettingEquation lpe = AgentInitializer.InitializeMetaCognitiveDecisionNetwork(lpm, ACSLevelProbabilitySettingEquation.Factory, Participant); lpe.Input.Add(honor.GetDriveStrength()); lpm.Commit(lpe); Participant.Commit(lpm); lpm.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 1; lpm.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 0; //Pre-train the IDN in the ACS PreTrainACS(idn); }
/// <summary> /// Setup the ACS subsystem /// </summary> private void SetupACS() { // Create Rule to avoid collision with wall SupportCalculator avoidCollisionWallSupportCalculator = FixedRuleToAvoidCollisionWall; FixedRule ruleAvoidCollisionWall = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputRotateClockwise, avoidCollisionWallSupportCalculator); // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleAvoidCollisionWall); // Create Rule To Go Ahead SupportCalculator goAheadSupportCalculator = FixedRuleToGoAhead; FixedRule ruleGoAhead = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputGoAhead, goAheadSupportCalculator); // Commit this rule to Agent (in the ACS) - FMT commenting to avoid conflict //CurrentAgent.Commit(ruleGoAhead); // FMT 29/04/2017 // FMT Create Rule to Eat SupportCalculator eatSupportCalculator = FixedRuleToEat; FixedRule ruleEat = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputEat, eatSupportCalculator); CurrentAgent.ACS.Parameters.PERFORM_RER_REFINEMENT = true; CurrentAgent.ACS.Parameters.LEVEL_SELECTION_METHOD = ActionCenteredSubsystem.LevelSelectionMethods.COMBINED; CurrentAgent.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; CurrentAgent.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 1; CurrentAgent.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 1; CurrentAgent.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 1; CurrentAgent.ACS.Parameters.WM_UPDATE_ACTION_PROBABILITY = 1; // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleEat); // FMT Create Rule to Get SupportCalculator getSupportCalculator = FixedRuleToGet; FixedRule ruleGet = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputGet, getSupportCalculator); // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleGet); // FMT Create Rule to Hide SupportCalculator hideSupportCalculator = FixedRuleToHide; FixedRule ruleHide = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputHide, hideSupportCalculator); // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleHide); // FMT Create Rule to Go To SupportCalculator gotoSupportCalculator = FixedRuleToGoTo; FixedRule ruleGoto = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputGoTo, gotoSupportCalculator); // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleGoto); // FMT Create Rule to Stop SupportCalculator stopSupportCalculator = FixedRuleToStop; FixedRule ruleStop = AgentInitializer.InitializeActionRule(CurrentAgent, FixedRule.Factory, outputStop, stopSupportCalculator); // Commit this rule to Agent (in the ACS) CurrentAgent.Commit(ruleStop); // Disable Rule Refinement CurrentAgent.ACS.Parameters.PERFORM_RER_REFINEMENT = false; // The selection type will be probabilistic CurrentAgent.ACS.Parameters.LEVEL_SELECTION_METHOD = ActionCenteredSubsystem.LevelSelectionMethods.STOCHASTIC; // The action selection will be fixed (not variable) i.e. only the statement defined above. CurrentAgent.ACS.Parameters.LEVEL_SELECTION_OPTION = ActionCenteredSubsystem.LevelSelectionOptions.FIXED; // Define Probabilistic values CurrentAgent.ACS.Parameters.FIXED_FR_LEVEL_SELECTION_MEASURE = 1; CurrentAgent.ACS.Parameters.FIXED_IRL_LEVEL_SELECTION_MEASURE = 0; CurrentAgent.ACS.Parameters.FIXED_BL_LEVEL_SELECTION_MEASURE = 0; CurrentAgent.ACS.Parameters.FIXED_RER_LEVEL_SELECTION_MEASURE = 0; // FMT 13/05/2017 additonal setting for network SimplifiedQBPNetwork net = AgentInitializer.InitializeImplicitDecisionNetwork(CurrentAgent, SimplifiedQBPNetwork.Factory); net.Parameters.LEARNING_RATE = 1; CurrentAgent.ACS.Parameters.PERFORM_RER_REFINEMENT = false; }
public void Initialize(Groups group) { World.Initialize(); John = World.NewAgent(); QBPNetwork idn = AgentInitializer.InitializeImplicitDecisionNetwork(John, QBPNetwork.Factory); World.NewDimensionValuePair("Target P", target); World.NewDimensionValuePair("Current P", target); World.NewExternalActionChunk(target); for (double i = 0; i < 12; i++) { if (World.GetDimensionValuePair("Target P", i) == null) { idn.Input.Add(World.NewDimensionValuePair("Target P", i)); idn.Input.Add(World.NewDimensionValuePair("Current P", i)); idn.Input.Add(World.NewExternalActionChunk(i)); idn.Output.Add(World.GetActionChunk(i)); } else { idn.Input.Add(World.GetDimensionValuePair("Target P", i)); idn.Input.Add(World.GetDimensionValuePair("Current P", i)); idn.Input.Add(World.GetActionChunk(i)); idn.Output.Add(World.GetActionChunk(i)); } } foreach (double i in As) { World.NewDimensionValuePair("A", i); } foreach (double i in Bs) { World.NewDimensionValuePair("B", i); } foreach (double i in Cs) { World.NewDimensionValuePair("C", i); } switch (group) { case Groups.VERBALIZATION: idn.Parameters.POSITIVE_MATCH_THRESHOLD = 1; RefineableActionRule.GlobalParameters.POSITIVE_MATCH_THRESHOLD = 1; RefineableActionRule.GlobalParameters.GENERALIZATION_THRESHOLD_1 = 1; RefineableActionRule.GlobalParameters.SPECIALIZATION_THRESHOLD_1 = .5; threshold_4 = .5; break; case Groups.MEMORY: for (double i = 0; i < 12; i++) { ExternalActionChunk w = (ExternalActionChunk)World.GetActionChunk((double)rand.Next(12)); var p = World.GetDimensionValuePair("Current P", FactoryOutput(i, (double)w.LabelAsIComparable)); ExternalActionChunk w1 = (ExternalActionChunk)World.GetActionChunk(Math.Round((target + p.Value + NoiseOptions[rand.Next(3)]) / 2)); FixedRule mfr = AgentInitializer.InitializeActionRule(John, FixedRule.Factory, w1, MemoryGroup_SupportCalculator); mfr.GeneralizedCondition.Add(p, true); mfr.GeneralizedCondition.Add(w, true); John.Commit(mfr); } goto default; case Groups.SIMPLE_RULE: for (double i = 0; i < 12; i++) { FixedRule sfr = AgentInitializer.InitializeActionRule(John, FixedRule.Factory, World.GetActionChunk(i), SimpleRule_SupportCalculator); John.Commit(sfr); } goto default; default: idn.Parameters.LEARNING_RATE = .05; idn.Parameters.DISCOUNT = .95; John.ACS.Parameters.SELECTION_TEMPERATURE = .09; idn.Parameters.POSITIVE_MATCH_THRESHOLD = 1; RefineableActionRule.GlobalParameters.GENERALIZATION_THRESHOLD_1 = 2; RefineableActionRule.GlobalParameters.SPECIALIZATION_THRESHOLD_1 = 1.2; RefineableActionRule.GlobalParameters.POSITIVE_MATCH_THRESHOLD = 1; threshold_4 = .2; break; } RefineableActionRule.GlobalParameters.INFORMATION_GAIN_OPTION = RefineableActionRule.IGOptions.PERFECT; John.Commit(idn); }
static void Main(string[] args) { //Initialize the task Console.WriteLine("Initializing the Full Hello World Task"); int CorrectCounter = 0; int NumberTrials = 20000; Random rand = new Random(); World.LoggingLevel = TraceLevel.Off; int progress = 0; TextWriter orig = Console.Out; StreamWriter sw = File.CreateText("HelloWorldFull.txt"); DimensionValuePair hi = World.NewDimensionValuePair("Salutation", "Hello"); DimensionValuePair bye = World.NewDimensionValuePair("Salutation", "Goodbye"); ExternalActionChunk sayHi = World.NewExternalActionChunk("Hello"); ExternalActionChunk sayBye = World.NewExternalActionChunk("Goodbye"); GoalChunk salute = World.NewGoalChunk("Salute"); GoalChunk bidFarewell = World.NewGoalChunk("Bid Farewell"); //Initialize the Agent Agent John = World.NewAgent("John"); SimplifiedQBPNetwork net = AgentInitializer.InitializeImplicitDecisionNetwork(John, SimplifiedQBPNetwork.Factory); net.Input.Add(salute, "goals"); net.Input.Add(bidFarewell, "goals"); net.Input.Add(hi); net.Input.Add(bye); net.Output.Add(sayHi); net.Output.Add(sayBye); net.Parameters.LEARNING_RATE = 1; John.Commit(net); John.ACS.Parameters.VARIABLE_BL_BETA = .5; John.ACS.Parameters.VARIABLE_RER_BETA = .5; John.ACS.Parameters.VARIABLE_IRL_BETA = 0; John.ACS.Parameters.VARIABLE_FR_BETA = 0; RefineableActionRule.GlobalParameters.SPECIALIZATION_THRESHOLD_1 = -.6; RefineableActionRule.GlobalParameters.GENERALIZATION_THRESHOLD_1 = -.1; RefineableActionRule.GlobalParameters.INFORMATION_GAIN_OPTION = RefineableActionRule.IGOptions.PERFECT; AffiliationBelongingnessDrive ab = AgentInitializer.InitializeDrive(John, AffiliationBelongingnessDrive.Factory, rand.NextDouble(), (DeficitChangeProcessor)HelloWorldFull_DeficitChange); DriveEquation abd = AgentInitializer.InitializeDriveComponent(ab, DriveEquation.Factory); ab.Commit(abd); John.Commit(ab); AutonomyDrive aut = AgentInitializer.InitializeDrive(John, AutonomyDrive.Factory, rand.NextDouble(), (DeficitChangeProcessor)HelloWorldFull_DeficitChange); DriveEquation autd = AgentInitializer.InitializeDriveComponent(aut, DriveEquation.Factory); aut.Commit(autd); John.Commit(aut); GoalSelectionModule gsm = AgentInitializer.InitializeMetaCognitiveModule(John, GoalSelectionModule.Factory); GoalSelectionEquation gse = AgentInitializer.InitializeMetaCognitiveDecisionNetwork(gsm, GoalSelectionEquation.Factory); gse.Input.Add(ab.GetDriveStrength()); gse.Input.Add(aut.GetDriveStrength()); GoalStructureUpdateActionChunk su = World.NewGoalStructureUpdateActionChunk(); GoalStructureUpdateActionChunk bu = World.NewGoalStructureUpdateActionChunk(); su.Add(GoalStructure.RecognizedActions.SET_RESET, salute); bu.Add(GoalStructure.RecognizedActions.SET_RESET, bidFarewell); gse.Output.Add(su); gse.Output.Add(bu); gsm.SetRelevance(su, ab, 1); gsm.SetRelevance(bu, aut, 1); gsm.Commit(gse); John.Commit(gsm); John.MS.Parameters.CURRENT_GOAL_ACTIVATION_OPTION = MotivationalSubsystem.CurrentGoalActivationOptions.FULL; //Run the task Console.WriteLine("Running the Full Hello World Task"); Console.SetOut(sw); SensoryInformation si; ExternalActionChunk chosen; for (int i = 0; i < NumberTrials; i++) { si = World.NewSensoryInformation(John); si[AffiliationBelongingnessDrive.MetaInfoReservations.STIMULUS, typeof(AffiliationBelongingnessDrive).Name] = 1; si[AutonomyDrive.MetaInfoReservations.STIMULUS, typeof(AutonomyDrive).Name] = 1; //Randomly choose an input to perceive. if (rand.NextDouble() < .5) { //Say "Hello" si.Add(hi, John.Parameters.MAX_ACTIVATION); si.Add(bye, John.Parameters.MIN_ACTIVATION); } else { //Say "Goodbye" si.Add(hi, John.Parameters.MIN_ACTIVATION); si.Add(bye, John.Parameters.MAX_ACTIVATION); } //Perceive the sensory information John.Perceive(si); //Choose an action chosen = John.GetChosenExternalAction(si); //Deliver appropriate feedback to the agent if (chosen == sayHi) { //The agent said "Hello". if (si[hi] == John.Parameters.MAX_ACTIVATION) { //The agent responded correctly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was correct"); //Record the agent's success. CorrectCounter++; //Give positive feedback. John.ReceiveFeedback(si, 1.0); } else { //The agent responded incorrectly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was incorrect"); //Give negative feedback. John.ReceiveFeedback(si, 0.0); } } else { //The agent said "Goodbye". if (si[bye] == John.Parameters.MAX_ACTIVATION) { //The agent responded correctly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was correct"); //Record the agent's success. CorrectCounter++; //Give positive feedback. John.ReceiveFeedback(si, 1.0); } else { //The agent responded incorrectly Trace.WriteLineIf(World.LoggingSwitch.TraceWarning, "John was incorrect"); //Give negative feedback. John.ReceiveFeedback(si, 0.0); } } Console.SetOut(orig); progress = (int)(((double)(i + 1) / (double)NumberTrials) * 100); Console.CursorLeft = 0; Console.Write(progress + "% Complete.."); Console.SetOut(sw); } //Report Results Console.WriteLine("Reporting Results for the Full Hello World Task"); Console.WriteLine("John got " + CorrectCounter + " correct out of " + NumberTrials + " trials (" + (int)Math.Round(((double)CorrectCounter / (double)NumberTrials) * 100) + "%)"); Console.WriteLine("At the end of the task, John had learned the following rules:"); foreach (var i in John.GetInternals(Agent.InternalContainers.ACTION_RULES)) { Console.WriteLine(i); } sw.Close(); Console.SetOut(orig); Console.CursorLeft = 0; Console.WriteLine("100% Complete.."); //Kill the agent to end the task Console.WriteLine("Killing John to end the program"); John.Die(); Console.WriteLine("John is Dead"); Console.WriteLine("The Full Hello World Task has finished"); Console.WriteLine("The results have been saved to \"HelloWorldFull.txt\""); Console.Write("Press any key to exit"); Console.ReadKey(true); }