def initial_pop(self):
print(f"{'=' * self.load_bar_len}\ncreating genesis population")
return [
Genome(i, self.network_params, self.mutation_scale,
self.w_mutation_rate, self.b_mutation_rate)
for i in range(self.population_size)
]
def initial_pop(self):
if self.verbose:
print('{0}\ncreating genisis population'.format(
'=' * self.verbose_load_bar))
genomes = []
for i in range(self.population_size):
genomes.append(
Genome(i, self.network_params, self.mutation_scale,
self.w_mutation_rate, self.b_mutation_rate))
return genomes
def evolve(self):
# genesis population
self.gen += 1
if self.gen == 1:
return
self.socket_reporter("genProgress", {"progress": 0})
print(f"{'=' * self.load_bar_len}\ncreating population {self.gen}")
# find fitness by normalizing score
self.normalize_score()
# find pool of genomes to breed and mutate
parents_1 = self.pool_selection()
parents_2 = self.pool_selection()
children = []
# create next generation
for idx, (p1, p2) in enumerate(zip(parents_1, parents_2)):
children.append(
Genome(idx,
self.network_params,
self.mutation_scale,
self.w_mutation_rate,
self.b_mutation_rate,
parent_1=self.genomes[p1],
parent_2=self.genomes[p2]
if np.random.random() < self.breeding_ratio else None))
progress = int((idx + 1) / len(parents_1) * self.load_bar_len)
progress_left = self.load_bar_len - progress
print('[{0}>{1}]'.format('=' * progress, ' ' * progress_left),
end='\r')
print(' ' * (self.load_bar_len + 3), end='\r')
self.genomes = children
# mutation scale will decay over time
self.mutation_scale *= self.mutation_decay
def evolve(self, g):
# genisis population
if g == 0:
return
if self.verbose:
print('{0}\ncreating population {1}'.format('='*self.verbose_load_bar, g+1))
# find fitness by normalizing score
self.normalize_score()
# find pool of genomes to breed and mutate
parents_1 = self.pool_selection()
parents_2 = self.pool_selection()
children = []
# evolving keras network
if self.network_params['network'] == 'convolutional':
# temporarily save models and clear session
configs, weights = [], []
for p1 in parents_1:
configs.append(self.genomes[p1].model.prediction.to_json())
weights.append(self.genomes[p1].model.prediction.get_weights())
tf.keras.backend.clear_session()
# reload models
for idx, (config, weight) in enumerate(zip(configs, weights)):
loaded = model_from_json(config)
loaded.set_weights(weight)
children.append(Genome(idx,
self.network_params,
self.mutation_scale,
self.w_mutation_rate,
self.b_mutation_rate,
load_keras=loaded))
if self.verbose:
progress = int((idx + 1)/len(parents_1) * self.verbose_load_bar)
progress_left = self.verbose_load_bar - progress
print('[{0}>{1}]'.format('=' * progress, ' ' * progress_left), end='\r')
else:
# create next generation
for idx, (p1, p2) in enumerate(zip(parents_1, parents_2)):
if np.random.random() < self.breeding_ratio:
# breeding
children.append(Genome(idx,
self.network_params,
self.mutation_scale,
self.w_mutation_rate,
self.b_mutation_rate,
parent_1=self.genomes[p1],
parent_2=self.genomes[p2]))
else:
# mutating
children.append(Genome(idx,
self.network_params,
self.mutation_scale,
self.w_mutation_rate,
self.b_mutation_rate,
parent_1=self.genomes[p1]))
if self.verbose:
progress = int((idx + 1)/len(parents_1) * self.verbose_load_bar)
progress_left = self.verbose_load_bar - progress
print('[{0}>{1}]'.format('=' * progress, ' ' * progress_left), end='\r')
if self.verbose: print(' ' * (self.verbose_load_bar + 3), end='\r')
self.genomes = children
# mutation scale will decay over time
self.mutation_scale *= self.mutation_decay