source: src/main/java/agents/anac/y2011/ValueModelAgent/ValueDecrease.java@ 126

package agents.anac.y2011.ValueModelAgent;

public class ValueDecrease{
        private double decrease;
        //well its not real std deviation, its
        //an abstraction, but is similar some what to weighted average
        private double avDev;
        private double reliabilty;
        private int lastSent;
        private double highestReliability;
        
        
        ValueDecrease(double val, double rel,double dev){
                avDev = dev;
                reliabilty = rel;
                decrease = val;
                lastSent=-1;
                highestReliability=rel;
        }
        public double getDecrease(){
                return decrease;
        }
        public void forceChangeDecrease(double newDecrease){
                decrease = newDecrease;
        }
        public double getReliabilty(){
                return reliabilty;
        }
        public double getMaxReliabilty(){
                return reliabilty;
        }
        public double getDeviance(){
                return avDev;
        }
        public int lastSent(){
                return lastSent;
        }
        public void sent(int bidIndex){
                lastSent = bidIndex;
        }
        //if the value is 100% reliable and the change is also very reliable
        //what chunk  of the value should be given to the new value
        //the actual effect of the new value is dependent on the reliabilty
        //of both.
        static private final double tempralDifferenceGamma = 0.1;
        private double sq(double x){return x*x;}
        public void updateWithNewValue(double newVal,double newReliability){
                if(reliabilty!=0.02){
                        double newChunk =  newReliability*(1-reliabilty);
                        double temporalC = reliabilty*tempralDifferenceGamma*newReliability;
                        newChunk+=temporalC;
                        double oldChunk =  reliabilty-temporalC;
                        double sumChunk = newChunk+oldChunk;
                        //newChunk/=sumChunk;
                        //oldChunk/=sumChunk;
                        double newDecrease = (newChunk*newVal+oldChunk*decrease)/sumChunk;
                        double change = Math.abs(newDecrease-decrease);
                        //this is a simplification of a real deviance calculation
                        double newDev = Math.sqrt((oldChunk/2*sq(avDev+change)
                                                        +oldChunk/2*sq(avDev-change)
                                                        +newChunk*sq(newVal-newDecrease))/sumChunk);
                        double temp = 1-change/(2*newDev);
                        reliabilty = oldChunk * (temp>0.2?temp:0.2);
                        temp = 1-(Math.abs(newDecrease-newVal)/(2*newDev));
                        reliabilty +=newChunk * (temp>0?temp:0);
                        decrease = newDecrease;
                        avDev = newDev;
                        if(highestReliability<reliabilty){
                                highestReliability = reliabilty;
                        }
                }
                //if this is the first time we really got this value than
                //the original value is meaningless
                else{
                        decrease = newVal;
                        reliabilty = newReliability;
                        avDev = 0.03;
                }
        }
}