public void runSupervised() {
MyTimer timer = new MyTimer();
timer.start();
PuddleMapFV agentfv = new PuddleMapFV(this.puddleMap, 5, 20, 20);
PuddleMapFVComponent agentCompFV = new PuddleMapFVComponent(this.puddleMap, 5, 20, 20);
StateToFeatureVectorGenerator svar =
new ConcatenatedObjectFeatureVectorGenerator(false, GridWorldDomain.CLASSAGENT);
java.util.List<EpisodeAnalysis> eas =
EpisodeAnalysis.parseFilesIntoEAList(this.expertDir, domain, this.sp);
LinearStateDifferentiableRF objectiveRF =
new LinearStateDifferentiableRF(agentfv, agentfv.getDim());
objectiveRF.setParameters(new double[] {1., -10, -10, 0, -10, 0, 0, 0, 0, 0});
WekaPolicy p = new WekaPolicy(agentfv, new J48(), this.domain.getActions(), eas);
// WekaPolicy p = new WekaPolicy(svar, new Logistic(), this.domain.getActions(), eas);
timer.stop();
System.out.println("Training Time: " + timer.getTime());
String baseName = "Svar";
GridWorldTerminalFunction tf = new GridWorldTerminalFunction(20, 20);
State simple = this.initialState.copy();
GridWorldDomain.setAgent(simple, 18, 0);
EpisodeAnalysis trainedEp1 = p.evaluateBehavior(simple, objectiveRF, tf, 200);
trainedEp1.writeToFile(trainedDir + "/j48" + baseName + "EpSimple", this.sp);
State hardAgent = this.initialState.copy();
GridWorldDomain.setAgent(hardAgent, 0, 9);
EpisodeAnalysis trainedEp2 = p.evaluateBehavior(hardAgent, objectiveRF, tf, 200);
trainedEp2.writeToFile(trainedDir + "/j48" + baseName + "EpHardAgent", this.sp);
int ngx = 12;
int ngy = 14;
tf = new GridWorldTerminalFunction(ngx, ngy);
this.puddleMap[ngx][ngy] = 1;
this.puddleMap[20][20] = 0;
agentfv.setGoal(ngx, ngy);
agentCompFV.setGoal(ngx, ngy);
State hardGoal = this.initialState.copy();
GridWorldDomain.setAgent(hardGoal, 0, 0);
EpisodeAnalysis trainedEp3 = p.evaluateBehavior(hardGoal, objectiveRF, tf, 200);
trainedEp3.writeToFile(trainedDir + "/j48" + baseName + "EpHardGoal", this.sp);
new EpisodeSequenceVisualizer(this.v, this.domain, this.sp, this.trainedDir);
}
