def runExperiment(architecture, dataset, seed, max_evaluations, num_samples):
from neuralbench.classification.dataset.create import run_validate_splits
np.random.seed(seed)
file_name = "backprop_%s_%s_%03d_e%010d_s%05d.dat" % (
architecture, dataset["name"], seed, max_evaluations, num_samples)
f = open(file_name, 'w')
f.write("seed\ttest_split\tvalidation_split")
f.write("\tevaluation_data\tevaluations\tfitness_type\tresult\n")
subs = createArchitecture(architecture, dataset["name"])
if dataset["name"] == "mnist":
train_network = configure_for_training(num_samples, max_evaluations,
10, subs, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
else:
train_network = configure_for_training(num_samples, max_evaluations, 2,
subs, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
f.close()
def runExperiment(architecture, dataset, seed, max_evaluations, num_samples):
from neuralbench.classification.dataset.create import run_validate_splits
params = LeeaParams()
params.parent_fitness_decay = 0.05
params.mutation_power_decay = 0.99
params.sexual_reproduction_proportion = 0.5
params.population_size = 50
params.starting_mutation_power = 0.3
params.mutation_power = params.starting_mutation_power
params.mutation_rate = 0.4
params.selection_proportion = 0.4
params.initial_weight_range = 4.
params.batch_size = num_samples
np.random.seed(seed)
eval_genotype, weights_template = leeaNetworks.createArchitecture(
architecture, dataset["name"])
np.random.seed(seed)
file_name = "leea_%s_%s_%03d_e%010d_s%05d.dat" % (
architecture, dataset["name"], seed, max_evaluations, num_samples)
f = open(file_name, 'w')
f.write("seed\ttest_split\tvalidation_split")
f.write(
"\tpopulation_size\tmutation_power\tmutation_rate\tselection_proportion\tbatch_size\tinitial_weight_range"
)
f.write("\tevaluation_data\tevaluations\tfitness_type\tresult\n")
if dataset["name"] == "mnist":
train_network = configure_for_training(params, max_evaluations, 10,
eval_genotype, weights_template,
seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
else:
train_network = configure_for_training(params, max_evaluations, 2,
eval_genotype, weights_template,
seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
f.close()
def runExperiment(architecture, dataset, seed, max_evaluations, num_samples):
params = CosyneParams()
params.population_size = 40
params.mutation_power = 0.3
params.mutation_rate = 0.4
params.selection_proportion = 0.4
params.initial_weight_range = 4.
params.batch_size = num_samples
np.random.seed(seed)
file_name = "cosyne_%s_%s_%03d_e%010d_s%05d.dat" % (
architecture, dataset["name"], seed, max_evaluations, num_samples)
f = open(file_name, 'w')
f.write("seed\ttest_split\tvalidation_split")
f.write(
"\tpopulation_size\tmutation_power\tmutation_rate\tselection_proportion\tbatch_size\tinitial_weight_range"
)
f.write("\tevaluation_data\tevaluations\tfitness_type\tresult\n")
buildNet, num_network_weights = cosyneNetworks.createArchitecture(
architecture, dataset["name"])
if dataset["name"] == "mnist":
from neuralbench.classification.dataset.create import run_validate_splits
eval_genotype = buildNet(28 * 28, 10)
train_network = configure_for_training(params, max_evaluations, 10,
eval_genotype,
num_network_weights, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
else:
from neuralbench.classification.dataset.create import run_validate_splits
eval_genotype = buildNet(2, 2)
train_network = configure_for_training(params, max_evaluations, 2,
eval_genotype,
num_network_weights, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
f.close()
def runExperiment(architecture,
dataset,
seed,
max_evaluations,
num_samples,
tanh=False):
from neuralbench.classification.dataset.create import run_validate_splits
np.random.seed(seed)
file_name = "hyperneat_%s_%s_%03d_e%010d_s%05d.dat" % (
architecture, dataset["name"], seed, max_evaluations, num_samples)
if tanh:
file_name = "hyperneattanh_%s_%s_%03d_e%010d_s%05d.dat" % (
architecture, dataset["name"], seed, max_evaluations, num_samples)
f = open(file_name, 'w')
subs = createArchitecture(architecture, dataset["name"])
set_activation_functions(tanh)
if dataset["name"] == "mnist":
train_network = configure_for_training(num_samples, max_evaluations,
10, subs, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
else:
train_network = configure_for_training(num_samples, max_evaluations, 2,
subs, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
f.close()
def runExperiment(dataset, seed, max_evaluations, num_samples, tanh=False):
from neuralbench.classification.dataset.create import run_validate_splits
np.random.seed(seed)
file_name = "neat_neat_%s_%03d_e%010d_s%05d.dat" % (
dataset["name"], seed, max_evaluations, num_samples)
if tanh:
file_name = "neattanh_neattanh_%s_%03d_e%010d_s%05d.dat" % (
dataset["name"], seed, max_evaluations, num_samples)
f = open(file_name, 'w')
f.write("seed\ttest_split\tvalidation_split")
f.write("\tevaluation_data\tevaluations\tfitness_type\tresult\n")
if tanh: set_activation_functions()
if dataset["name"] == "mnist":
train_network = configure_for_training(num_samples, max_evaluations,
10, seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
else:
train_network = configure_for_training(num_samples, max_evaluations, 2,
seed, f)
run_validate_splits(train_network,
dataset["X_train"],
dataset["y_train"],
dataset["X_test"],
dataset["y_test"],
folds=10,
seed=seed)
f.close()