mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
# Normalize and reshape data
x_train, x_test = x_train / 255.0, x_test / 255.0
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)
# Create dataset object, which controls all the data
normalized_dataset = Dataset(
training_examples=x_train,
training_labels=y_train,
testing_examples=x_test,
testing_labels=y_test,
validation_split=0.1,
)
# Create backend responsible for training & validating
backend = TFKerasBackend(dataset=normalized_dataset)
# Create DeepSwarm object responsible for optimization
deepswarm = DeepSwarm(backend=backend)
# Find the topology for a given dataset
try:
topology = deepswarm.find_topology()
except:
Log.error(f'{sys.exc_info()} occured')
Log.error(f'{traceback.format_exc()}')
# Evaluate discovered topology
deepswarm.evaluate_topology(topology)
# Train topology for additional 30 epochs
trained_topology = deepswarm.train_topology(topology, 30)
# Evaluate the final topology
deepswarm.evaluate_topology(trained_topology)
# Load CIFAR-10 dataset
cifar10 = tf.keras.datasets.cifar10
(x_train, y_train), (x_test, y_test) = cifar10.load_data()
# Convert class vectors to binary class matrices
y_train = tf.keras.utils.to_categorical(y_train, 10)
y_test = tf.keras.utils.to_categorical(y_test, 10)
# Create dataset object, which controls all the data
dataset = Dataset(
training_examples=x_train,
training_labels=y_train,
testing_examples=x_test,
testing_labels=y_test,
validation_split=0.1,
)
# Create backend responsible for training & validating
backend = TFKerasBackend(dataset=dataset)
# Create DeepSwarm object responsible for optimization
deepswarm = DeepSwarm(backend=backend)
# Find the topology for a given dataset
topology = deepswarm.find_topology()
# Evaluate discovered topology
deepswarm.evaluate_topology(topology)
# Train topology on augmented data for additional 50 epochs
trained_topology = deepswarm.train_topology(topology,
50,
augment={
'rotation_range': 15,
'width_shift_range': 0.1,
'height_shift_range': 0.1,
'horizontal_flip': True,
})
def aco_train(self, ui):
# function which uses aco to train on the inputted dataset
global x_train, y_train, x_test, y_test, trained_topology
# modify YAML file by taking a parameter
if ui.dataset == 'cifar10':
inp_fo = open("settings/cifar10.yaml").read() # Read the Yaml File
yaml = YAML() # Load the yaml object
# print(ui.m_depth, ui.n_ant, ui.epochs)
code = yaml.load(
inp_fo) # Load content of YAML file to yaml object
code['DeepSwarm'][
'max_depth'] = ui.m_depth # Update Yaml file with new parameter in object
code['DeepSwarm']['aco'][
'ant_count'] = ui.n_ant # Update Yaml file with new parameter in object
code['DeepSwarm']['backend'][
'epochs'] = ui.epochs # Update Yaml file with new parameter in object
inp_fo2 = open("settings/cifar10.yaml",
"w") # Open the file for Write
yaml.dump(code, inp_fo2) # Write to file with new parameter
inp_fo2.close() # close the file
elif ui.dataset == 'fashion-mnist':
inp_fo = open(
"settings/fashion-mnist.yaml").read() # Read the Yaml File
yaml = YAML() # Load the yaml object
# print(ui.m_depth, ui.n_ant, ui.epochs)
code = yaml.load(
inp_fo) # Load content of YAML file to yaml object
code['DeepSwarm'][
'max_depth'] = ui.m_depth # Update Yaml file with new parameter in object
code['DeepSwarm']['aco'][
'ant_count'] = ui.n_ant # Update Yaml file with new parameter in object
code['DeepSwarm']['backend'][
'epochs'] = ui.epochs # Update Yaml file with new parameter in object
inp_fo2 = open("settings/fashion-mnist.yaml",
"w") # Open the file for Write
yaml.dump(code, inp_fo2) # Write to file with new parameter
inp_fo2.close() # close the file
# important to leave deepswarm import at this point otherwise
# user customization won't update the yaml file
from deepswarm.backends import Dataset, TFKerasBackend
print("ACO Training of dataset ...")
if ui.dataset == 'fashion-mnist':
(x_train, y_train), (x_test, y_test) = fashion_mnist.load_data(
) # Load Fashion MNIST dataset
x_train, x_test = x_train / 255.0, x_test / 255.0 # Normalize and reshape data
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
x_test = x_test.reshape(x_test.shape[0], 28, 28, 1)
elif ui.dataset == 'cifar10':
(x_train,
y_train), (x_test,
y_test) = cifar10.load_data() # Load CIFAR-10 dataset
y_train = tf.keras.utils.to_categorical(
y_train, 10) # Convert class vectors to binary class matrices
y_test = tf.keras.utils.to_categorical(y_test, 10)
# import is at this point since _init_.py for deepswarm module is invoked once module is imported
# the modified yaml settings (above) will not be loaded if deepswarm imported at start of prorgram.
from deepswarm.deepswarm import DeepSwarm
# Create dataset object, which controls all the data
dataset = Dataset(training_examples=x_train,
training_labels=y_train,
testing_examples=x_test,
testing_labels=y_test,
validation_split=0.1)
backend = TFKerasBackend(
dataset=dataset
) # Create backend responsible for training & validating
deepswarm = DeepSwarm(
backend=backend
) # Create DeepSwarm object responsible for optimization
topology = deepswarm.find_topology(
) # Find the topology for a given dataset
deepswarm.evaluate_topology(topology) # Evaluate discovered topology
if ui.dataset == 'cifar10':
trained_topology = deepswarm.train_topology(
topology,
50,
augment={
'rotation_range': 15,
'width_shift_range': 0.1,
'height_shift_range': 0.1,
'horizontal_flip': True,
})
if ui.dataset == 'fashion-mnist':
# Train topology on augmented data for additional 50 epochs
trained_topology = deepswarm.train_topology(
topology,
50,
augment={
'rotation_range': 15,
'width_shift_range': 0.1,
'height_shift_range': 0.1,
'horizontal_flip': True,
})
deepswarm.evaluate_topology(
trained_topology) # Evaluate the final topology
return