Add files via upload

src/hk/edu/gaSchedule/algorithm/Cso.py

@@ -0,0 +1,168 @@
+from .NsgaIII import NsgaIII
+import math
+import numpy as np
+import random
+from time import time
+
+
+# X. -S. Yang and Suash Deb, "Cuckoo Search via Lévy flights,"
+# 2009 World Congress on Nature & Biologically Inspired Computing (NaBIC), Coimbatore, India,
+# 2009, pp. 210-214, doi: 10.1109/NABIC.2009.5393690.
+# Copyright (c) 2023 - 2024 Miller Cy Chan
+
+
+# Cuckoo Search Optimization (CSO)
+class Cso(NsgaIII):
+    def __init__(self, configuration, numberOfCrossoverPoints=2, mutationSize=2, crossoverProbability=80,
+                 mutationProbability=3, maxIterations=5000):
+        self._max_iterations = maxIterations
+        self._maxRepeat = min(15, self._max_iterations // 2)
+        super().__init__(configuration, numberOfCrossoverPoints, mutationSize, crossoverProbability,
+                        mutationProbability)
+
+        # there should be at least 5 chromosomes in population
+        if self._populationSize < 5:
+            self._populationSize = 5
+
+        self._chromlen, self._pa, self._beta = 0, .25, 1.5
+        num = math.gamma(1 + self._beta) * math.sin(math.pi * self._beta / 2)
+        den = math.gamma((1 + self._beta) / 2) * self._beta * (2 ** ((self._beta - 1) / 2))
+        self._σu, self._σv = (num / den) ** (1 / self._beta), 1
+
+        self._gBestScore = None
+        self._current_position = [[]]
+
+
+    def initialize(self, population):
+        prototype = self._prototype
+
+        populationSize = len(population)
+        for i in range(populationSize):
+            positions = []
+            # add new chromosome to population
+            population[i] = prototype.makeNewFromPrototype(positions)
+            if i < 1:
+                self._chromlen = len(positions)
+                self._current_position = np.zeros((populationSize, self._chromlen), dtype=float)
+
+
+    def optimum(self, localVal, chromosome):
+        localBest = self._prototype.makeEmptyFromPrototype()
+        localBest.updatePositions(localVal)
+
+        if localBest.dominates(chromosome):
+            chromosome.updatePositions(localVal)
+            return localVal
+
+        positions = np.zeros(self._chromlen, dtype=float)
+        chromosome.extractPositions(positions)
+        return positions
+
+
+    def updatePosition1(self, population):
+        current_position = np.copy(self._current_position)
+        populationSize = self._populationSize
+        u, v = np.random.randn(populationSize) * self._σu, np.random.randn(populationSize) * self._σv
+        S = u / (np.abs(v) ** (1 / self._beta))
+
+        for i in range(populationSize):
+            if self._gBestScore is None:
+                self._gBestScore = np.zeros(self._chromlen, dtype=float)
+                population[i].extractPositions(self._gBestScore)
+            else:
+                self._gBestScore = self.optimum(self._gBestScore, population[i])
+
+            self._current_position[i] += np.random.normal(self._chromlen) * 0.01 * S[i] * (current_position[i] - self._gBestScore)
+            self._current_position[i] = self.optimum(self._current_position[i], population[i])
+
+
+    def updatePosition2(self, population):
+        current_position = np.copy(self._current_position)
+        populationSize = self._populationSize
+        for i in range(populationSize):
+            d1, d2 = np.random.randint(0, 5, 2)
+            changed = False
+            for j in range(self._chromlen):
+                r = np.random.rand()
+                if r < self._pa:
+                    changed = True
+                    self._current_position[i, j] += random.random() * (current_position[d1, j] - current_position[d2, j])
+
+            if changed:
+                self._current_position[i] = self.optimum(self._current_position[i], population[i])
+
+
+    def reform(self):
+        random.seed(round(time() * 1000))
+        np.random.seed(int(time()))
+        if self._crossoverProbability < 95:
+            self._crossoverProbability += 1.0
+        elif self._pa < .5:
+            self._pa += .01
+
+
+    def replacement(self, population):
+        populationSize = self._populationSize
+        self.updatePosition1(population)
+        self.updatePosition2(population)
+
+        for i in range(populationSize):
+            chromosome = self._prototype.makeEmptyFromPrototype()
+            chromosome.updatePositions(self._current_position[i])
+            population[i] = chromosome
+
+        return super().replacement(population)
+
+
+    # Starts and executes algorithm
+    def run(self, maxRepeat=9999, minFitness=0.999):
+        mutationSize, mutationProbability = self._mutationSize, self._mutationProbability
+        populationSize = self._populationSize
+        population = populationSize * [None]
+
+        self.initialize(population)
+        random.seed(round(time() * 1000))
+        np.random.seed(int(time()))
+        pop = [population, None]
+
+        # Current generation
+        currentGeneration = 0
+        bestNotEnhance, lastBestFit = 0, 0.0
+
+        cur, next = 0, 1
+        while currentGeneration < self._max_iterations:
+            if currentGeneration > 0:
+                best = self.result
+                print("Fitness:", "{:f}\t".format(best.fitness), "Generation:", currentGeneration, end="\r")
+
+                # algorithm has reached criteria?
+                if best.fitness > minFitness:
+                    break
+
+                difference = abs(best.fitness - lastBestFit)
+                if difference <= 0.0000001:
+                    bestNotEnhance += 1
+                else:
+                    lastBestFit = best.fitness
+                    bestNotEnhance = 0
+
+                if bestNotEnhance > (maxRepeat / 50):
+                    self.reform()
+
+            # crossover
+            offspring = self.crossing(pop[cur])
+
+            # mutation
+            [i for i in map(self.mutation, offspring)]
+
+            pop[cur].extend(offspring)
+
+            # replacement
+            pop[next] = self.replacement(pop[cur])
+            self._best = pop[next][0] if pop[next][0].dominates(pop[cur][0]) else pop[cur][0]
+
+            cur, next = next, cur
+            currentGeneration += 1
+
+    def __str__(self):
+        return "Cuckoo Search Optimization (CSO)"