package ch.idsia.ai.agents.ai;
import ch.idsia.ai.MLP;
import ch.idsia.ai.Evolvable;
import ch.idsia.ai.agents.Agent;
import ch.idsia.ai.agents.RegisterableAgent;
import ch.idsia.mario.environments.Environment;
/**
* Created by IntelliJ IDEA.
* User: julian
* Date: Jun 14, 2009
* Time: 1:43:04 PM
*/
public class LargeMLPAgent extends RegisterableAgent implements Agent, Evolvable {
static private String name = "LargeMLPAgent";
private MLP mlp;
final int numberOfOutputs = Environment.numberOfButtons;
final int numberOfInputs = 101;
public LargeMLPAgent() {
super (name);
mlp = new MLP (numberOfInputs, 10, numberOfOutputs);
}
private LargeMLPAgent(MLP mlp) {
super (name);
this.mlp = mlp;
}
public Evolvable getNewInstance() {
return new LargeMLPAgent(mlp.getNewInstance());
}
public Evolvable copy() {
return new LargeMLPAgent(mlp.copy ());
}
public void reset() {
mlp.reset ();
}
public void mutate() {
mlp.mutate ();
}
public boolean[] getAction(Environment observation) {
double[] inputs;// = new double[numberOfInputs];
byte[][] scene = observation.getLevelSceneObservation(/*1*/);
byte[][] enemies = observation.getEnemiesObservation(/*0*/);
inputs = new double[numberOfInputs];
int which = 0;
for (int i = -3; i < 4; i++) {
for (int j = -3; j < 4; j++) {
inputs[which++] = probe(i, j, scene);
}
}
for (int i = -3; i < 4; i++) {
for (int j = -3; j < 4; j++) {
inputs[which++] = probe(i, j, enemies);
}
}
inputs[inputs.length - 3] = observation.isMarioOnGround() ? 1 : 0;
inputs[inputs.length - 2] = observation.mayMarioJump() ? 1 : 0;
inputs[inputs.length - 1] = 1;
double[] outputs = mlp.propagate (inputs);
boolean[] action = new boolean[numberOfOutputs];
for (int i = 0; i < action.length; i++) {
action[i] = outputs[i] > 0;
}
return action;
}
public Agent.AGENT_TYPE getType() {
return Agent.AGENT_TYPE.AI;
}
public String getName() {
return name;
}
public void setName(String name) {
}
private double probe (int x, int y, byte[][] scene) {
int realX = x + 11;
int realY = y + 11;
return (scene[realX][realY] != 0) ? 1 : 0;
}
}