source: framspy/evolalg/base/experiment_abc.py

Last change on this file was 1294, checked in by Maciej Komosinski, 10 months ago

Initialize more fields in constructors

File size: 11.0 KB
Line 
1import argparse
2import json
3import os
4import pickle
5import time
6import math
7import random
8from abc import ABC, abstractmethod
9
10from ..base.random_sequence_index import RandomIndexSequence
11from ..constants import BAD_FITNESS
12from ..json_encoders import Encoder
13from ..structures.hall_of_fame import HallOfFame
14from ..structures.individual import Individual
15from ..structures.population import PopulationStructures
16
17
18class ExperimentABC(ABC):
19
20    def __init__(self, popsize, hof_size, save_only_best) -> None:
21        self.population_structures = None
22        self.stats = []
23        self.current_generation = 0
24        self.time_elapsed = 0
25
26        self.hof = HallOfFame(hof_size)
27        self.popsize = popsize
28        self.save_only_best = save_only_best
29
30    def select(self, individuals, tournament_size, random_index_sequence):
31        """Tournament selection, returns the index of the best individual from those taking part in the tournament"""
32        best_index = None
33        for i in range(tournament_size):
34            rnd_index = random_index_sequence.getNext()
35            if best_index is None or individuals[rnd_index].fitness > best_index.fitness:
36                best_index = individuals[rnd_index]
37        return best_index
38
39    def addGenotypeIfValid(self, ind_list, genotype):
40        new_individual = Individual()
41        new_individual.set_and_evaluate(genotype, self.evaluate)
42        if new_individual.fitness is not BAD_FITNESS:
43            ind_list.append(new_individual)
44
45    @staticmethod
46    def stochastic_round(value): # https://en.wikipedia.org/wiki/Rounding#Stochastic_rounding
47        # For example, value==2.1 should turn in most cases to int 2, rarely to int 3
48        lower = math.floor(value) # returns int
49        return lower + (random.random() < (value - lower))
50
51    def make_new_population(self, individuals, prob_mut, prob_xov, tournament_size):
52        """'individuals' is the input population (a list of individuals).
53        Assumptions: all genotypes in 'individuals' are valid and evaluated (have fitness set).
54        Returns: a new population of size 'self.popsize' with prob_mut mutants, prob_xov offspring, and the remainder of clones."""
55
56        # if (self.popsize * probability) below is not integer (e.g. popsize=50, prob_xov=0.333, expected number of crossovers = 50*0.333=16.65), stochastic_round() will ensure that you will get on average the expected number of crossovers per generation (e.g. 16.65: less often 16, more often 17).
57        expected_mut = self.stochastic_round(self.popsize * prob_mut) # or int(...) if you accept less mutants in some cases, see the comment above
58        expected_xov = self.stochastic_round(self.popsize * prob_xov) # or int(...) if you accept less crossovers in some cases, see the comment above
59        assert expected_mut + expected_xov <= self.popsize, "If probabilities of mutation (%g) and crossover (%g) added together exceed 1.0, then the population would grow every generation..." % (prob_mut, prob_xov) # can be triggered due to stochastic_round() if prob_mut+prob_xov is close to 1 and the expected number of mutants or crossovers is not integer; if this happens, adjust popsize, prob_mut and prob_xov accordingly.
60
61        newpop = []
62        ris = RandomIndexSequence(len(individuals))
63
64        # adding valid mutants of selected individuals...
65        while len(newpop) < expected_mut:
66            ind = self.select(individuals, tournament_size=tournament_size, random_index_sequence=ris)
67            self.addGenotypeIfValid(newpop, self.mutate(ind.genotype))
68
69        # adding valid crossovers of selected individuals...
70        while len(newpop) < expected_mut + expected_xov:
71            ind1 = self.select(individuals, tournament_size=tournament_size, random_index_sequence=ris)
72            ind2 = self.select(individuals, tournament_size=tournament_size, random_index_sequence=ris)
73            self.addGenotypeIfValid(newpop, self.cross_over(ind1.genotype, ind2.genotype))
74
75        # select clones to fill up the new population until we reach the same size as the input population
76        while len(newpop) < self.popsize:
77            ind = self.select(individuals, tournament_size=tournament_size, random_index_sequence=ris)
78            newpop.append(Individual().copyFrom(ind))
79
80        return newpop
81
82    def save_state(self, state_filename):
83        state = self.get_state()
84        if state_filename is None:
85            return
86        state_filename_tmp = state_filename + ".tmp"
87        try:
88            with open(state_filename_tmp, "wb") as f:
89                pickle.dump(state, f)
90            # ensures the new file was first saved OK (e.g. enough free space on device), then replace
91            os.replace(state_filename_tmp, state_filename)
92        except Exception as ex:
93            raise RuntimeError("Failed to save evolution state '%s' (because: %s). This does not prevent the experiment from continuing, but let's stop here to fix the problem with saving state files." % (
94                state_filename_tmp, ex))
95
96    def load_state(self, state_filename):
97        if state_filename is None:
98            # print("Loading evolution state: file name not provided")
99            return None
100        try:
101            with open(state_filename, 'rb') as f:
102                state = pickle.load(f)
103                self.set_state(state)
104        except FileNotFoundError:
105            return None
106        print("...Loaded evolution state from '%s'" % state_filename)
107        return True
108
109    def get_state_filename(self, save_file_name):
110        return None if save_file_name is None else save_file_name + '_state.pkl'
111
112    def get_state(self):
113        return [self.time_elapsed, self.current_generation, self.population_structures, self.hof, self.stats]
114
115    def set_state(self, state):
116        self.time_elapsed, self.current_generation, self.population_structures, hof_, self.stats = state
117        # sorting: ensure that we add from worst to best so all individuals are added to HOF
118        for h in sorted(hof_, key=lambda x: x.rawfitness):
119            self.hof.add(h)
120
121    def update_stats(self, generation, all_individuals):
122        worst = min(all_individuals, key=lambda item: item.rawfitness)
123        best = max(all_individuals, key=lambda item: item.rawfitness)
124        # instead of single best, could add all individuals in population here, but then the outcome would depend on the order of adding
125        self.hof.add(best)
126        self.stats.append(best.rawfitness if self.save_only_best else best)
127        print("%d\t%d\t%g\t%g" % (generation, len(all_individuals), worst.rawfitness, best.rawfitness))
128
129    def initialize_evolution(self, initialgenotype):
130        self.current_generation = 0
131        self.time_elapsed = 0
132        self.stats = []  # stores the best individuals, one from each generation
133        initial_individual = Individual()
134        initial_individual.set_and_evaluate(initialgenotype, self.evaluate)
135        self.hof.add(initial_individual)
136        self.stats.append(
137            initial_individual.rawfitness if self.save_only_best else initial_individual)
138        self.population_structures = PopulationStructures(
139            initial_individual=initial_individual, archive_size=0, popsize=self.popsize)
140
141    def evolve(self, hof_savefile, generations, initialgenotype, pmut, pxov, tournament_size):
142        file_name = self.get_state_filename(hof_savefile)
143        state = self.load_state(file_name)
144        if state is not None:  # loaded state from file
145            # saved generation has been completed, start with the next one
146            self.current_generation += 1
147            print("...Resuming from saved state: population size = %d, hof size = %d, stats size = %d, archive size = %d, generation = %d/%d" % (len(self.population_structures.population), len(self.hof), len(self.stats), (len(self.population_structures.archive)), self.current_generation, generations))  # self.current_generation (and g) are 0-based, parsed_args.generations is 1-based
148        else:
149            self.initialize_evolution(initialgenotype)
150        time0 = time.process_time()
151        for g in range(self.current_generation, generations):
152            self.population_structures.population = self.make_new_population(
153                self.population_structures.population, pmut, pxov, tournament_size)
154            self.update_stats(g, self.population_structures.population)
155            if hof_savefile is not None:
156                self.current_generation = g
157                self.time_elapsed += time.process_time() - time0
158                self.save_state(file_name)
159        if hof_savefile is not None:
160            self.save_genotypes(hof_savefile)
161        return self.population_structures.population, self.stats
162
163    @abstractmethod
164    def mutate(self, gen1):
165        pass
166
167    @abstractmethod
168    def cross_over(self, gen1, gen2):
169        pass
170
171    @abstractmethod
172    def evaluate(self, genotype):
173        pass
174
175    def save_genotypes(self, filename):
176        """Implement if you want to save finall genotypes,in default implementation this function is run once at the end of evolution"""
177        state_to_save = {
178            "number_of_generations": self.current_generation,
179            "hof": [{"genotype": individual.genotype,
180                     "fitness": individual.rawfitness} for individual in self.hof.hof]}
181        with open(f"{filename}.json", 'w') as f:
182            json.dump(state_to_save, f, cls=Encoder)
183
184   
185    @staticmethod
186    def get_args_for_parser():
187        parser = argparse.ArgumentParser()
188        parser.add_argument('-popsize', type=int, default=50,
189                            help="Population size, default: 50.")
190        parser.add_argument('-generations', type=int, default=5,
191                                help="Number of generations, default: 5.")
192        parser.add_argument('-tournament', type=int, default=5,
193                            help="Tournament size, default: 5.")
194        parser.add_argument('-pmut', type=float, default=0.7,
195                        help="Probability of mutation, default: 0.7")
196        parser.add_argument('-pxov', type=float, default=0.2,
197                        help="Probability of crossover, default: 0.2")
198        parser.add_argument('-hof_size', type=int, default=10,
199                            help="Number of genotypes in Hall of Fame. Default: 10.")
200        parser.add_argument('-hof_savefile', type=str, required=False,
201                                help= 'If set, Hall of Fame will be saved in Framsticks file format (recommended extension *.gen. This also activates saving state (checpoint} file and auto-resuming from the saved state, if this file exists.')
202        parser.add_argument('-save_only_best', type=bool, default=True, required=False,
203                            help="")
204        parser.add_argument('-fitness_set_negative_to_zero', action='store_true',
205                            help="This flag forces fitness to become max(0,fitness), so it is always made non-negative. Using niching or novelty techniques without this flag (thus allowing negative fitness values) requires verifying/improving fitness diversification formulas to work as intended for both positive and negative fitness values.")
206       
207        return parser
Note: See TracBrowser for help on using the repository browser.