public static FastKNN loadKNN(BTFData btf) throws IOException {
System.out.println("[AntKNN] Loading BTF data...");
FastKNN knn = new FastKNN(FEATURE_DIM, 3);
String[] desiredVel = btf.loadColumn("dvel");
String[] desiredVelBool = btf.loadColumn("dbool");
String[] wallVec = btf.loadColumn("wallvec");
String[] wallBool = btf.loadColumn("wallbool");
String[] antVec = btf.loadColumn("antvec");
String[] antBool = btf.loadColumn("antbool");
String[] homeVec = btf.loadColumn("homevec");
String[] foodVec = btf.loadColumn("foodvec");
String[] prevVec = btf.loadColumn("pvel");
String[] prevBoolVec = btf.loadColumn("pbool");
int numRows = desiredVel.length;
double[] sample = new double[FEATURE_DIM];
double[] velclass = new double[3];
for (int i = 0; i < numRows; i++) {
if (i % (numRows / 10) == 0) System.out.println("[AntKNN] " + i + "/" + numRows);
if (Boolean.parseBoolean(desiredVelBool[i])) {
String[] tmp = desiredVel[i].split(" ");
velclass[0] = Double.parseDouble(tmp[0]);
velclass[1] = Double.parseDouble(tmp[1]);
velclass[2] = Double.parseDouble(tmp[2]);
tmp = antVec[i].split(" ");
sample[0] = Double.parseDouble(tmp[0]);
sample[1] = Double.parseDouble(tmp[1]);
tmp = wallVec[i].split(" ");
sample[2] = Double.parseDouble(tmp[0]);
sample[3] = Double.parseDouble(tmp[1]);
tmp = homeVec[i].split(" ");
sample[4] = Double.parseDouble(tmp[0]);
sample[5] = Double.parseDouble(tmp[1]);
// tmp = foodVec[i].split(" ");
// sample[6] = Double.parseDouble(tmp[0]);
// sample[7] = Double.parseDouble(tmp[1]);
// tmp = prevVec[i].split(" ");
// sample[4] = Double.parseDouble(tmp[0]);
// sample[5] = Double.parseDouble(tmp[1]);
// sample[6] = Double.parseDouble(tmp[2]);
// tmp = homeVec[i].split(" ");
// sample[7] = Double.parseDouble(tmp[0]);
// sample[8] = Double.parseDouble(tmp[1]);
knn.add(sample, velclass);
}
}
// sigmaNormalize currently broken, don't use it! (Dec 4th, 2012)
// knn.sigmaNormalize();
System.out.println("[AntKNN] Done!");
return knn;
}
public void act(double time) {
// System.out.println("Ant body:"+antBody);
double[] rv = new double[3];
MutableDouble2D ant = new MutableDouble2D();
boolean sawAnt = antBody.getNearestSameTypeVec(ant);
MutableDouble2D wall = new MutableDouble2D();
boolean sawWall = antBody.getNearestObstacleVec(wall);
MutableDouble2D home = new MutableDouble2D();
boolean sawHome = antBody.getPoiDir(home, "nest");
MutableDouble2D food = new MutableDouble2D();
boolean sawFood = antBody.getNearestPreyVec(food);
double[] sensorVec = new double[FEATURE_DIM];
double[][] nearestK = new double[5][3];
sensorVec[0] = ant.x;
sensorVec[1] = ant.y;
sensorVec[2] = wall.x;
sensorVec[3] = wall.y;
// System.out.println("Sensor vec: ["+sensorVec[0]+", "+sensorVec[1]+", "+sensorVec[2]+",
// "+sensorVec[3]+"]");
sensorVec[4] = home.x;
sensorVec[5] = home.y;
// sensorVec[6] = food.x;
// sensorVec[7] = food.y;
// sensorVec[4] = prevVel[0];
// sensorVec[5] = prevVel[1];
// sensorVec[6] = prevVel[2];
// sensorVec[7] = home.x;
// sensorVec[8] = home.y;
knn.query(sensorVec, nearestK);
// now, do median, average, or random selection
// average
for (int i = 0; i < rv.length; i++) rv[i] = 0.0;
for (int i = 0; i < nearestK.length; i++) {
for (int j = 0; j < nearestK[i].length; j++) {
rv[j] += nearestK[i][j];
}
}
for (int i = 0; i < rv.length; i++) rv[i] = rv[i] / (double) nearestK.length;
// if(rv[0] < 0 ) System.out.println("Movin backwards!");
prevVel[0] = rv[0];
prevVel[1] = rv[1];
prevVel[2] = rv[2];
// System.out.println("rv: ["+rv[0]+", "+rv[1]+", "+rv[2]+"]");
antBody.setDesiredVelocity(rv[0], rv[1], rv[2]);
}