def evolve(n, debugging=False):
if (debugging):
debug = open("debug.txt", "w")
else:
debug = None
config.load('configCIFAR10')
# Create 2 separate populations (size is now defined explicitly, but config file can still be used)
module_pop = population.Population(15,
chromosome.ModuleChromo,
debug=debug)
# As the top hierarchical level, the blueprint population needs to be able to see the module population
blueprint_pop = population.Population(10,
chromosome.BlueprintChromo,
module_pop,
debug=debug)
# Most of the actual evolving is now handled outside of the population, by CoDeepNEAT
# Instead of requiring the user to overwrite the evaluation function, CoDeepNEAT evaluates the populations itself,
# it simply requires a fitness function for the networks it creates passed in as an argument.
codeepneat.epoch(n,
blueprint_pop,
module_pop,
25,
fitness,
data,
save_best=True,
name='CIFAR10',
debug=debug)
# It will still stop if fitness surpasses the max_fitness_threshold in config file
# Plots the evolution of the best/average fitness
visualize.plot_stats(module_pop.stats, name="CIFAR10mod_")
visualize.plot_stats(blueprint_pop.stats, name="CIFAR10bp_")
def mutate_module(chromo_file):
fp = open(chromo_file, "rb")
#print(fp)
chromo = pickle.load(fp)
#print(chromo)
fp.close()
print("module:" + chromo_file)
for connection in chromo._connections:
print(str(connection))
counter = 0
config.load('configCIFAR10')
inputs = keras.layers.Input(config.Config.input_nodes, name='input')
module = chromo.decode(inputs)
plot_model(module,
to_file='mutate/module' + str(counter) + '.png',
show_shapes=True)
while True:
counter += 1
mutate = input("Mutate? y/n: ")
print()
if mutate == "y":
chromo._mutate_add_layer_debug()
for connection in chromo._connections:
print(str(connection))
config.load('configCIFAR10')
inputs = keras.layers.Input(config.Config.input_nodes,
name='input')
module = chromo.decode(inputs)
plot_model(module,
to_file='mutate/module' + str(counter) + '.png',
show_shapes=True)
else:
return
def eval_best(model_file, data):
config.load('configCIFAR10')
model = keras.models.load_model(model_file)
datagen = ImageDataGenerator(
featurewise_center=False, # set input mean to 0 over the dataset
samplewise_center=False, # set each sample mean to 0
featurewise_std_normalization=
False, # divide inputs by std of the dataset
samplewise_std_normalization=False, # divide each input by its std
zca_whitening=False, # apply ZCA whitening
rotation_range=
10, # randomly rotate images in the range (degrees, 0 to 180)
width_shift_range=
0.1, # randomly shift images horizontally (fraction of total width)
height_shift_range=
0.1, # randomly shift images vertically (fraction of total height)
horizontal_flip=True, # randomly flip images
vertical_flip=False) # randomly flip images
# Compute quantities required for featurewise normalization
# (std, mean, and principal components if ZCA whitening is applied).
datagen.fit(data[0])
csv_logger = CSVLogger(model_file + '_cifar10_new.csv')
# Fit the model on the batches generated by datagen.flow().
model.fit_generator(datagen.flow(data[0], data[1], batch_size=batch_size),
steps_per_epoch=data[0].shape[0] // batch_size,
validation_data=(data[2], data[3]),
epochs=epoch,
verbose=1,
max_queue_size=100,
callbacks=[lr_reducer, early_stopper, csv_logger])
loss, fitness = model.evaluate(data[2], data[3])
print("fitness", fitness)
def visualize_module(chromo_file):
#config.load('configCoverage')
#chromosome.node_gene_type = genome.NodeGene
fp = open(chromo_file, "rb")
#print(fp)
chromo = pickle.load(fp)
print(chromo)
fp.close()
config.load('../configCIFAR10')
inputs = keras.layers.Input(config.Config.input_nodes, name='input')
module = chromo.decode(inputs)
plot_model(module, to_file='module.png', show_shapes = True)