Paru le 16/11/2013 | Relié
Tout public
package rlpark.plugin.rltoys.algorithms.predictions.td;
import rlpark.plugin.rltoys.algorithms.traces.ATraces;
import rlpark.plugin.rltoys.algorithms.traces.EligibilityTraceAlgorithm;
import rlpark.plugin.rltoys.algorithms.traces.Traces;
import rlpark.plugin.rltoys.math.vector.RealVector;
import rlpark.plugin.core.api.monitoring.annotations.Monitor;
public class TDLambda extends TD implements EligibilityTraceAlgorithm
{
private static final long serialVersionUID = 8613865620293286722L;
protected final double lambda;
@Monitor
public final Traces e;
double gamma_t;
public TDLambda(double lambda, double gamma, double alpha, int nbFeatures) {
this(lambda, gamma, alpha, nbFeatures, new ATraces());
}
public TDLambda(double lambda, double gamma, double alpha, int
nbFeatures, Traces prototype) {
super(gamma, alpha, nbFeatures);
this.lambda = lambda;
e = prototype.newTraces(nbFeatures);
}
@Override
protected double initEpisode() {
e.clear();
gamma_t = gamma;
return super.initEpisode();
}
@Override
public double update(RealVector x_t, RealVector x_tp1, double r_ tp1, double gamma_tp1) {
if (x_t == null)
return initEpisode()
v_t = v.dotProduct(x_t);
delta_t = r_tp1 + gamma_tp1 * v.dotProduct(x_tp1) - v_t;
e.update(lambda * gamma_t, x_t);
v.addToSelf(alpha_v * delta_t, e.vect());
gamma_t = gamma_tp1;
return delta_t;
}
@Override
public void resetWeight(int index) {
super.resetWeight(index);
e.vect().setEntry(index, 0);
}
@Override
public Traces traces() {
return e
}}
