import java.io.*; import java.util.*; public class PredictionGAwithPittLCS { public int popsize; private int series_length; private int chrom_length; private int rule_length; private int num_rules_per_chrom; public double[] fitnesses; private char [] binarySeries; public PredictionGAwithPittLCS(int pop_size_input, int series_length_input, int chrom_length_input, int rule_length_input) { popsize=pop_size_input; fitnesses = new double[popsize]; series_length = series_length_input; binarySeries = new char[series_length]; chrom_length = chrom_length_input; rule_length= rule_length_input; num_rules_per_chrom = chrom_length/rule_length; } private void readFile() throws Exception { String s; int i,j,line_length; j= 0; BufferedReader input = new BufferedReader(new FileReader("BTSPdata.txt")); s = input.readLine(); while(s != null) { line_length = s.length(); for(i = 0; i < line_length; i++) binarySeries[j++] = s.charAt(i); s = input.readLine(); } input.close(); } public void objective(char chromosomes[][]) { char[] chromosome = new char[chrom_length]; // new for CS 326: I made a char array // to store a single // chromosome, which needs to be evaluated. for(int j = 0; j < popsize; j++) // Go through pop { for(int i = 0; i < chrom_length; i++) // CS 326 This loop gets the j_th chromosome from chromosome[i] = chromosomes[i][j]; // CS 326 the population of chromosomes, and stores // them in the char array "chromosome" for // you to hash on. (chromosome is the "key" for table.) /* if(Hashtable.contains(chromosome)) // CS 326: THIS IS PSEUDOCODE! Gives basic idea of how fitnesses[j] = (double) Hashtable.get(chromosome); // to check hashtable first, to see if // the j_th chromosome is in there. If // hashtable contains j_th chromosome, then // just get the fitness (which is the "value" // for the table) and store it in the j_th // spot in the fitness array. else // Otherwise, if j_th chromosome NOT in hashtable, { // then must evaluate it using the fitness function // (code below) AND, then, don't forget to "put" // the newly computed fitness into the hashtable, // paired with the chromosome. (See comments 30 // lines below. */ fitnesses[j] = 0.0; for(int k = 0; k < series_length-rule_length; k++) // Go through time series (bit array) { int num_votes_for_a_one = 0; // Determine the rule group's decision for(int r = 0; r < num_rules_per_chrom; r++) // by checking what each rule in the group "votes" (0, or 1) { int i = r*rule_length; // For each rule in the group while(i < (r+1)*rule_length // Compare the rule to the current set of bits from the series... && (chromosomes[i][j]=='#' // See if the rule matches... '#'s match anything || chromosomes[i][j] == binarySeries[k+(i-r*rule_length)])) i++; if(i == (r+1)*rule_length) // If this is true, then rule r matched the input bits num_votes_for_a_one++; // which means rule r votes for a "1" as next bit } // Implement Nate's "Squeaky Wheel Rule", which means if any one rule in group (chrom) votes for a "one" // then they all go along with that (the group/chrom predicts a one for the next bit). Otherwise (if // no rule votes for a one), then the group votes for a zero as the next bit. if((num_votes_for_a_one >0 && binarySeries[k+rule_length] == '1') // If any rule in chrom. voted for a one, // then the whole chrom. votes for a one, and then if || // next bit really IS a one, then incr. chrom's fitness. // OR, if no rule in chrom. voted for a one, then the whole (num_votes_for_a_one == 0 && binarySeries[k+rule_length] == '0')) // chrom votes for a zero, and then if the next // bit really IS a zero, then incr. chrom.'s fitness. fitnesses[j] += 1.0; } /* and put (chromosome, fitnesses[j]) into Hashtable. // CS 326 PSEUDOCODE! Put the new // fitness into hashtable, paired with // the chromosome. */ } } public static void main(String args[]) throws Exception { final int CHROM_LENGTH = 64; // CHROM_LENGTH = RULE_LENGTH * number of rules in chrom final int RULE_LENGTH =8; // so make sure CHROM_LENGTH is a multiple of RULE_LENGTH! final int SERIES_LENGTH = 10000; // this is how many bits are used in obj. fcn. final int POPSIZE= 1000; // Population size final long RANDOM_SEED = 1567890; Random randomizer= new Random(RANDOM_SEED); final int MAX_NUM_GENS = 5000; // SIMPLE GA PARAMETERS final double PROB_CROSSOVER= 0.9; // for each two parents, prob. children will be xover of parents final double PROB_MUTATE = 0.00; // bit-wise prob. of bit-flip final int tourn_size = 2; final double DONT_CARE_PROBABILITY = 0.5; // Prob. of a '#' per bit position in // each chromosome of each individual in // initial population (gen 0). double worst, best, avg; // Run stats int index_of_best; int gen_num = 0; int competitor_1; int competitor_2; int winner; int parent_1, parent_2; int crossover_pt_1, crossover_pt_2; int temp; // Open files for writing fitness stats each generation PrintStream best_of_gen_file = new PrintStream(new FileOutputStream("best_of_gen.txt")); PrintStream avg_of_gen_file = new PrintStream(new FileOutputStream("avg_of_gen.txt")); PrintStream worst_of_gen_file = new PrintStream(new FileOutputStream("worst_of_gen.txt")); // Create BB simulator. Create chrom array. PredictionGAwithPittLCS predictor = new PredictionGAwithPittLCS(POPSIZE, SERIES_LENGTH, CHROM_LENGTH, RULE_LENGTH); predictor.readFile(); char pop[][]= new char[CHROM_LENGTH][POPSIZE]; // Current pop of bit string chroms // Generate initial pop. randomly for(int j=0; j< POPSIZE;j++) for (int i = 0;i < CHROM_LENGTH;i++) if(randomizer.nextDouble()< DONT_CARE_PROBABILITY) pop[i][j] = '#'; else if(randomizer.nextBoolean()) pop[i][j] = '0'; else pop[i][j] = '1'; // Generation loop while(gen_num < MAX_NUM_GENS) { // evaluate population predictor.objective(pop); // Print population to console /* System.out.print("pop = "); for(int j=0;j best) { best = predictor.fitnesses[j]; index_of_best = j; } if(predictor.fitnesses[j] < worst) worst = predictor.fitnesses[j]; avg += predictor.fitnesses[j]; } avg = avg/POPSIZE; System.out.println("------------" + gen_num +"---------------------"); System.out.print("Best design in pop: " + best + " "); for(int h = 0; h < CHROM_LENGTH; h++) if(h%RULE_LENGTH == 0) System.out.print(" "+pop[h][index_of_best]); else System.out.print(""+pop[h][index_of_best]); System.out.println(); System.out.println("Avg. cost of designs: " + avg); System.out.println("Worst design in pop: " + worst); System.out.println(); // Print fitness stats to fitness stats files best_of_gen_file.println(best); avg_of_gen_file.println(avg); worst_of_gen_file.println(worst); char new_pop[][] = new char[CHROM_LENGTH][POPSIZE]; // Selection for(int j = 0; j < POPSIZE; j++) { competitor_1 = randomizer.nextInt(POPSIZE); competitor_2 = randomizer.nextInt(POPSIZE); if(predictor.fitnesses[competitor_1] > predictor.fitnesses[competitor_2]) winner = competitor_1; else winner = competitor_2; for(int i = 0; i < CHROM_LENGTH; i++) new_pop[i][j] = pop[i][winner]; } pop = new_pop; // CROSSOVER (two point) if(PROB_CROSSOVER > 0.0) // if crossover turned off, skip it! { new_pop = new char[CHROM_LENGTH][POPSIZE]; for(int j = 0; j < POPSIZE; j+=2) { parent_1 = randomizer.nextInt(POPSIZE); parent_2 = randomizer.nextInt(POPSIZE); if(randomizer.nextDouble() > PROB_CROSSOVER) { crossover_pt_1 = CHROM_LENGTH; crossover_pt_2 = CHROM_LENGTH; } else { crossover_pt_1 = randomizer.nextInt(CHROM_LENGTH); crossover_pt_2 = randomizer.nextInt(CHROM_LENGTH); if(crossover_pt_2 < crossover_pt_1) { temp = crossover_pt_1; crossover_pt_1 = crossover_pt_2; crossover_pt_2 = temp; } } for(int i=0; i < crossover_pt_1; i++) { new_pop[i][j] = pop[i][parent_1]; new_pop[i][j+1] = pop[i][parent_2]; } for(int i=crossover_pt_1; i < crossover_pt_2; i++) { new_pop[i][j] = pop[i][parent_2]; new_pop[i][j+1] = pop[i][parent_1]; } for(int i=crossover_pt_2; i < CHROM_LENGTH; i++) { new_pop[i][j] = pop[i][parent_1]; new_pop[i][j+1] = pop[i][parent_2]; } } pop = new_pop; } // END CROSSOVER // MUTATION if(PROB_MUTATE > 0.0) // if mutation turned off, skip this part { new_pop = new char[CHROM_LENGTH][POPSIZE]; for(int j = 0; j < POPSIZE; j++) // For each indiv. for(int i=0; i PROB_MUTATE) // bit-wise prob. of mutation new_pop[i][j] = pop[i][j]; else if (pop[i][j] == '#') if(randomizer.nextBoolean()) new_pop[i][j] = '0'; else new_pop[i][j] = '1'; else if(randomizer.nextBoolean()) new_pop[i][j] = '#'; else if(pop[i][j] == '0') new_pop[i][j]='1'; else new_pop[i][j]= '0'; pop = new_pop; } // END MUTATION gen_num++; } // END OF MAIN GENERATION while LOOP // Close fitness stats files best_of_gen_file.println(); avg_of_gen_file.flush(); worst_of_gen_file.close(); System.exit(0); // end of program, end of run } }