/**                                      
 *
 * --Copyright notice-- 
 *
 * Author: Andy Evans, based on code by Dimitris Ballas.
 *
 * This code is also licensed under GPL2 which can be found at 
 * the Open Source Initiative website at... 
 * http://cran.r-project.org/web/licenses/GPL-2
 *
 * --End of Copyright notice-- 
 *
**/
// Simulated Annealling code from simple Microsimulation App. You can find the app online at:
// http://www.geog.leeds.ac.uk/groups/mass/internal/software


/**
 * The meat of this particular program. Takes each area in turn, and swaps people in and out until statistics ok.<P>
 * The code works broadly by swapping random people out of the area, and replacing them by
 * another random person. The new error between the current statistics and those we're hoping
 * for is assessed, and if there's an improvement the change is kept, otherwise the old person is
 * put back in and the new one removed. This "gradient descent" style method is adjusted by the
 * Simulated Annealing algorithm, which allows worse errors to be kept with a probablity that
 * reduces over time. As this is a toy application with only one
 * attribute of two values, the SA routine actually slows down the basic gradient
 * descent algorithm, but if there were multiple attributes that needed fitting, it would
 * be a real boon.
**/

private void redistribute() {
	
	for (int area = 0; area < numberOfAreas; area++) {
	    
	    messageBox.append("\n\nDoing area " + area + "\n");
	    
	    // Set up the SA temperature to drop. 
	    
	    double temperature = 0;
	    int minError = -1;
	    Vector minErrorPeople = null;
	    int areaError = 0;
	    
	    for (int i = maxTemperature; i > 0; i--) {
		
		// The next line is lifted almost entirely from Dimitris' SimLeeds.
		
		temperature = (double)temperatureConversion*((double)i/(double)maxTemperature); 
		
		// Calculate the current error for the area. We're going
		// to carry on until the error is low, or we exceed a fixed
		// number of runs. In addition, each time a minimum error set 
		// of people is generated, we record this, as the algorithm 
		// can wander off into bad solutions and get lost.
		
		areaError = calculateError(area);
		if (minError == -1) {
		    minError = areaError;
		    minErrorPeople = (Vector)(world[area].clone());
		} 
		int runs = 0;
		
		// Start swapping.
		
		while ((areaError > errorMargin) && (runs < maxRuns)) {
		    
		    // Replace one of the people in the area with someone new.
		    
		    int person = (int)(world[area].size() * Math.random());
		    Person oldPerson = (Person) world[area].elementAt(person);
		    world[area].remove(oldPerson);
		    Person newPerson = getRandomPerson();
		    world[area].addElement(newPerson);

		    // Recalculate the error and decide whether to keep them or not.
		    
		    int newAreaError = calculateError(area);
		    if (newAreaError > areaError) {
			
			// Keep bad choices with a probablity relating to how bad they are and 
			// the current temperature.
			
			// The next line is lifted almost entirely from Dimitris' SimLeeds.
			
			if (Math.random() < Math.exp((-1 * ((double)newAreaError - (double)areaError))/temperature)) {
			    areaError = newAreaError;
			} else {
			    world[area].remove(newPerson);
			    world[area].addElement(oldPerson);
			}

		    } else {
			areaError = newAreaError;
			
			// If this is the lowest error we've seen, record the people.
			if (areaError < minError) {
			    minError = areaError;
			    minErrorPeople = (Vector)(world[area].clone());
			}
		    }
		    
		    messageBox.append(tableToReplicate.rowToString(area) + " Target " + currentTable.rowToString(area) + 
			" error = " + areaError + "\n");
		    runs++;
		    
		} // End of swapping while loop.
		
		// If we're ok with the current answer, don't bother reducing the temperature.
		
		if (areaError < errorMargin) break;
		
	    } // End of temperature decrease for loop.

	    // Check to make sure the error is the lowest we've seen, and if it isn't, replace the people.
	    
	    if (areaError > minError) {
		world[area] = minErrorPeople;
	    }
	    
	} // End of doing all the areas.
	
	messageBox.append("\n\nDone:\n");
	for (int area = 0; area < numberOfAreas; area++) {
	    messageBox.append(tableToReplicate.rowToString(area) + " Target " + currentTable.rowToString(area) + "\n");
	}
	
    } // End of redistribute.