refactor algorithms to not deal with cumulative fit

README.md

@@ -33,6 +33,8 @@ git clone https://github.com/schoonhovenrichard/BlooPy.git
 - bitarray
 - pyswarms
 - networkx
+- pytest
+- hypothesis
 
 ## Implemented algorithms
 ### Discrete local search algorithms

bloopy/algorithms/discrete_diffevo.py

@@ -135,7 +135,6 @@ def create_offspring2(self, parents):
             else:
                 bprime.fitness = self.ffunc(bprime.bitstring)
                 self.visited_cache[bsstr] = bprime.fitness
-                self.cumulative_fit += bprime.fitness
                 self.func_evals += 1
 
             # If better, replace candidate
@@ -227,7 +226,6 @@ def create_offspring(self, parents):
             else:
                 bprime.fitness = self.ffunc(bprime.bitstring)
                 self.visited_cache[bsstr] = bprime.fitness
-                self.cumulative_fit += bprime.fitness
                 self.func_evals += 1
 
             # If better, replace candidate

bloopy/algorithms/genetic_algorithm.py

@@ -64,7 +64,6 @@ class and overriding the 'point_mutate' method.
             self.boundary_list = utils.generate_boundary_list(searchspace)
         self.visited_cache = dict()
         self.func_evals = 0
-        self.cumulative_fit = 0
         if mutation is not None:
             self.mutation = max(1, int(self.bs_size * mutation))
         else:
@@ -90,7 +89,6 @@ def set_fit_pop(self, pop):
             else:
                 pop[i].fitness = self.ffunc(pop[i].bitstring)
                 self.visited_cache[bsstr] = pop[i].fitness
-                self.cumulative_fit += pop[i].fitness
                 self.func_evals += 1
 
     def generate_random_pop(self):
@@ -105,7 +103,6 @@ def generate_random_pop(self):
                 new_specimen.fitness = self.visited_cache[bsstr]
             else:
                 new_specimen.fitness = self.ffunc(new_specimen.bitstring)
-                self.cumulative_fit += new_specimen.fitness
                 self.func_evals += 1
                 self.visited_cache[bsstr] = new_specimen.fitness
             self.current_pop.append(new_specimen)
@@ -221,8 +218,6 @@ def solve(self,
             best_gen_so_far (int): Generation when best was found.
             variance (float):  Termination variance of population.
             self.func_evals (int): Total number of Fevals performed.
-            self.cumulative_fit (float): Cumulative fitness of all
-                fitness function evaluations.
         """
         generation = 0
         best_fit = self.current_best().fitness
@@ -277,4 +272,4 @@ def solve(self,
 
         if verbose:
             print("Terminated after {0} generations with best fitness: {1:.3f} | # of fitness evals: {2}".format(generation, best_fit, self.func_evals))
-        return (best_fit, self.current_best(), best_gen_so_far, variance, self.func_evals, self.cumulative_fit)
+        return (best_fit, self.current_best(), best_gen_so_far, variance, self.func_evals)

bloopy/algorithms/genetic_local_search.py

@@ -68,12 +68,10 @@ def one_generation(self):
             else:
                 children[i].fitness = self.ffunc(children[i].bitstring)
                 self.visited_cache[bsstr] = children[i].fitness
-                self.cumulative_fit += children[i].fitness
                 self.func_evals += 1
 
-            children[i], feval, tot_fit, self.visited_cache = self.hillclimber(children[i], self.ffunc, self.minmax, self.func_evals, self.maxfeval, self.visited_cache, self.nbour_method)
+            children[i], feval, self.visited_cache = self.hillclimber(children[i], self.ffunc, self.minmax, self.func_evals, self.maxfeval, self.visited_cache, self.nbour_method)
             self.func_evals += feval
-            self.cumulative_fit += tot_fit
 
         # Selection step
         self.current_pop = self.selector(parents, children, self.minmax)

bloopy/algorithms/hillclimbers.py

@@ -12,7 +12,6 @@ def RandomGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited
     splits = candidate.boundary_list
     foundimprove = True
     func_evals = 0
-    total_fit = 0
     bs_size = candidate.size
     while foundimprove:
         child = individual(candidate.size, bitstring=copy.deepcopy(candidate.bitstring), boundary_list=splits)
@@ -34,7 +33,6 @@ def RandomGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited
                     child.fitness = ffunc(child.bitstring)
                     visited_cache[bsstr] = child.fitness
                     func_evals += 1
-                    total_fit += child.fitness
                 if minmax * child.fitness > minmax * candidate.fitness:
                     candidate = copy.deepcopy(child)
                     foundimprove = True
@@ -71,7 +69,6 @@ def RandomGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited
                         child.fitness = ffunc(child.bitstring)
                         visited_cache[bsstr] = child.fitness
                         func_evals += 1
-                        total_fit += child.fitness
                     if minmax * child.fitness > minmax * candidate.fitness:
                         candidate = copy.deepcopy(child)
                         indices[k] = i
@@ -84,15 +81,14 @@ def RandomGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited
                         child.fitness = candidate.fitness
                 if foundimprove and restart:
                     break
-    return candidate, func_evals, total_fit, visited_cache
+    return candidate, func_evals, visited_cache
 
 def OrderedGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache, nbour_method, order, restart=True):
     if nbour_method not in ["Hamming", "adjacent"]:
         raise Exception("Unknown neighbour method.")
     splits = candidate.boundary_list
     foundimprove = True
     func_evals = 0
-    total_fit = 0
     bs_size = candidate.size
     while foundimprove:
         child = individual(candidate.size, bitstring=copy.deepcopy(candidate.bitstring), boundary_list=splits)
@@ -115,7 +111,6 @@ def OrderedGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visite
                     child.fitness = ffunc(child.bitstring)
                     visited_cache[bsstr] = child.fitness
                     func_evals += 1
-                    total_fit += child.fitness
                 if minmax * child.fitness > minmax * candidate.fitness:
                     candidate = copy.deepcopy(child)
                     foundimprove = True
@@ -154,7 +149,6 @@ def OrderedGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visite
                         child.fitness = ffunc(child.bitstring)
                         visited_cache[bsstr] = child.fitness
                         func_evals += 1
-                        total_fit += child.fitness
                     if minmax * child.fitness > minmax * candidate.fitness:
                         candidate = copy.deepcopy(child)
                         indices[k] = i
@@ -167,15 +161,14 @@ def OrderedGreedyHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visite
                         child.fitness = candidate.fitness
                 if foundimprove and restart:
                     break
-    return candidate, func_evals, total_fit, visited_cache
+    return candidate, func_evals, visited_cache
 
 def BestHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache, nbour_method):
     if nbour_method not in ["Hamming", "adjacent"]:
         raise Exception("Unknown neighbour method.")
     splits = candidate.boundary_list
     foundimprove = True
     func_evals = 0
-    total_fit = 0
     bs_size = candidate.size
     while foundimprove:
         child = individual(candidate.size, bitstring=copy.deepcopy(candidate.bitstring), boundary_list=splits)
@@ -196,7 +189,6 @@ def BestHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache,
                     child.fitness = ffunc(child.bitstring)
                     visited_cache[bsstr] = child.fitness
                     func_evals += 1
-                    total_fit += child.fitness
 
                 #If a neighbour is better, than the best found neighbours, save it
                 if minmax * child.fitness > minmax * best_child.fitness:
@@ -236,7 +228,6 @@ def BestHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache,
                         child.fitness = ffunc(child.bitstring)
                         visited_cache[bsstr] = child.fitness
                         func_evals += 1
-                        total_fit += child.fitness
 
                     #If a neighbour is better, than the best found neighbours, save it
                     if minmax * child.fitness > minmax * best_child.fitness:
@@ -250,15 +241,14 @@ def BestHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache,
             if minmax * best_child.fitness > minmax * candidate.fitness:
                 foundimprove = True
                 candidate = copy.deepcopy(best_child)
-    return candidate, func_evals, total_fit, visited_cache
+    return candidate, func_evals, visited_cache
 
 def StochasticHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_cache, nbour_method):
     if nbour_method not in ["Hamming", "adjacent"]:
         raise Exception("Unknown neighbour method.")
     splits = candidate.boundary_list
     foundimprove = True
     func_evals = 0
-    total_fit = 0
     bs_size = candidate.size
     while foundimprove:
         if maxfeval is not None and totfevals >= maxfeval:
@@ -280,7 +270,6 @@ def StochasticHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_c
                     child.fitness = ffunc(child.bitstring)
                     visited_cache[bsstr] = child.fitness
                     func_evals += 1
-                    total_fit += child.fitness
 
                 if minmax * child.fitness > minmax * candidate.fitness:
                     improved_candidate = copy.deepcopy(child)
@@ -314,7 +303,6 @@ def StochasticHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_c
                         child.fitness = ffunc(child.bitstring)
                         visited_cache[bsstr] = child.fitness
                         func_evals += 1
-                        total_fit += child.fitness
                     if minmax * child.fitness > minmax * candidate.fitness:
                         improved_candidate = copy.deepcopy(child)
                         uphill_moves.append((improved_candidate, abs(child.fitness-candidate.fitness)))
@@ -328,4 +316,4 @@ def StochasticHillclimb(candidate, ffunc, minmax, totfevals, maxfeval, visited_c
             # Choose a random uphill move proportionate to fitness increase
             index = np.random.choice(range(len(uphill_moves)),size=(1,),p=(np.array(probs)/float(sum(probs))).tolist())[0]
             candidate = copy.deepcopy(uphill_moves[index][0])
-    return candidate, func_evals, total_fit, visited_cache
+    return candidate, func_evals, visited_cache

bloopy/algorithms/iterative_local_search.py

@@ -59,7 +59,6 @@ def randomwalk(self, maxfeval):
                 self.current_candidate.fitness = self.ffunc(self.current_candidate.bitstring)
                 self.visited_cache[bsstr] = self.current_candidate.fitness
                 self.func_evals += 1
-                self.cumulative_fit += self.current_candidate.fitness
         else:
             self.current_candidate = individual(self.best_candidate.size, bitstring=copy.deepcopy(self.best_candidate.bitstring), boundary_list=self.boundary_list)
             self.current_candidate.fitness = self.best_candidate.fitness
@@ -73,7 +72,6 @@ def randomwalk(self, maxfeval):
                 self.current_candidate.fitness = self.ffunc(self.current_candidate.bitstring)
                 self.visited_cache[bsstr] = self.current_candidate.fitness
                 self.func_evals += 1
-                self.cumulative_fit += self.current_candidate.fitness
 
 class BestILS(iterative_local_search_base):
     def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk, noimprove=100, searchspace=None, neighbour='Hamming'):
@@ -97,9 +95,8 @@ def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk
         super().__init__(fitness_function, bitstring_size, minmax_problem, random_walk, noimprove, searchspace=searchspace, neighbour=neighbour)
 
     def hillclimb_candidate(self, maxfeval):
-        self.current_candidate, extra_fevals, totfit, self.visited_cache = hillclimb.BestHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method)
+        self.current_candidate, extra_fevals, self.visited_cache = hillclimb.BestHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method)
         self.func_evals += extra_fevals
-        self.cumulative_fit += totfit
 
 class RandomGreedyILS(iterative_local_search_base):
     def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk, noimprove=100, searchspace=None, neighbour='Hamming', restart_search=True):
@@ -127,9 +124,8 @@ def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk
         self.restart = restart_search
 
     def hillclimb_candidate(self, maxfeval):
-        self.current_candidate, extra_fevals, totfit, self.visited_cache = hillclimb.RandomGreedyHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method, restart=self.restart)
+        self.current_candidate, extra_fevals, self.visited_cache = hillclimb.RandomGreedyHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method, restart=self.restart)
         self.func_evals += extra_fevals
-        self.cumulative_fit += totfit
 
 class OrderedGreedyILS(iterative_local_search_base):
     def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk, noimprove=100, searchspace=None, neighbour='Hamming', restart_search=True, order=None):
@@ -160,9 +156,8 @@ def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk
         self.order = order
 
     def hillclimb_candidate(self, maxfeval):
-        self.current_candidate, extra_fevals, totfit, self.visited_cache = hillclimb.OrderedGreedyHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method, self.order, restart=self.restart)
+        self.current_candidate, extra_fevals, self.visited_cache = hillclimb.OrderedGreedyHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method, self.order, restart=self.restart)
         self.func_evals += extra_fevals
-        self.cumulative_fit += totfit
 
 class StochasticILS(iterative_local_search_base):
     def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk, noimprove=100, searchspace=None, neighbour='Hamming'):
@@ -188,6 +183,5 @@ def __init__(self, fitness_function, bitstring_size, minmax_problem, random_walk
         super().__init__(fitness_function, bitstring_size, minmax_problem, random_walk, noimprove, searchspace=searchspace, neighbour=neighbour)
 
     def hillclimb_candidate(self, maxfeval):
-        self.current_candidate, extra_fevals, totfit, self.visited_cache = hillclimb.StochasticHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method)
+        self.current_candidate, extra_fevals, self.visited_cache = hillclimb.StochasticHillclimb(self.current_candidate, self.ffunc, self.minmax, self.func_evals, maxfeval, self.visited_cache, self.nbour_method)
         self.func_evals += extra_fevals
-        self.cumulative_fit += totfit