normalization='batch',
id='softmax')
optimizer_settings = {
'args': {
'momentum': 0.9
},
'initial_lr': 0.1,
'optimizer': 'SGD'
}
solver = MXSolver(
batch_size=64,
devices=(args.gpu_index, ),
epochs=30,
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
from data_utilities import load_mnist
data = load_mnist(path='stretched_canvas_mnist', scale=1,
shape=(1, 56, 56))[:2]
data += load_mnist(path='stretched_mnist', scale=1, shape=(1, 56, 56))[2:]
info = solver.train(data)
postfix = '-' + args.postfix if args.postfix else ''
identifier = 'residual-network-on-stretched-mnist-%d%s' % (
args.n_residual_layers, postfix)
network = nin(activate)
lr = 0.1
lr_table = {100000 : lr * 0.1}
lr_scheduler = AtIterationScheduler(lr, lr_table)
optimizer_settings = {
'args' : {'momentum' : 0.9},
'initial_lr' : lr,
'lr_scheduler' : lr_scheduler,
'optimizer' : 'SGD',
'weight_decay' : 0.0001,
}
solver = MXSolver(
batch_size = BATCH_SIZE,
devices = (0, 1, 2, 3),
epochs = 300,
initializer = DReLUInitializer(0, 1),
optimizer_settings = optimizer_settings,
symbol = network,
verbose = True,
)
data = load_cifar10_record(BATCH_SIZE)
info = solver.train(data)
identifier = 'residual-network-n-%d-activate-%s-times-%d' % (N, ACTIVATE, TIMES)
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters = solver.export_parameters()
pickle.dump(parameters, open('parameters/%s' % identifier, 'wb'))
'args': {
'momentum': 0.9
},
'initial_lr': args.initial_lr,
'lr_scheduler': lr_scheduler,
'optimizer': 'SGD',
'weight_decay': 0.0001,
}
from mx_solver import MXSolver
from mx_initializer import PReLUInitializer
solver = MXSolver(
batch_size=args.batch_size,
devices=(0, 1, 2, 3),
epochs=150,
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
from data_utilities import load_cifar10_record
data = load_cifar10_record(args.batch_size)
info = solver.train(data)
postfix = '-' + args.postfix if args.postfix else ''
identifier = 'rnn-attention-network-on-cifar-10-%d%s' % (args.n_layers,
postfix)
import cPickle as pickle
lr_scheduler = AtIterationScheduler(lr, lr_table)
optimizer_settings = {
'args': {
'momentum': 0.9
},
'initial_lr': lr,
'lr_scheduler': lr_scheduler,
'optimizer': 'SGD',
}
solver = MXSolver(
batch_size=BATCH_SIZE,
devices=GPU_availability()[:1],
epochs=150,
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
data = load_cifar10(center=True, rescale=True)
# data = load_cifar10_record(BATCH_SIZE)
info = solver.train(data)
identifier = 'dropping-out-mlp-%d-%d' % (N_LAYERS, N_HIDDEN_UNITS)
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters = solver.export_parameters()
pickle.dump(parameters, open('parameters/%s' % identifier, 'wb'))
network = layers.fully_connected(X=network, n_hidden_units=10)
network = layers.softmax_loss(prediction=network, normalization='batch', id='softmax')
BATCH_SIZE = 128
lr = 0.1
lr_table = {32000 : lr * 0.1, 48000 : lr * 0.01}
lr_scheduler = AtIterationScheduler(lr, lr_table)
optimizer_settings = {
'args' : {'momentum' : 0.9},
'initial_lr' : lr,
'lr_scheduler' : lr_scheduler,
'optimizer' : 'SGD',
'weight_decay' : 0.0001,
}
solver = MXSolver(
batch_size = BATCH_SIZE,
devices = (0, 1, 2, 3),
epochs = 150,
initializer = PReLUInitializer(),
optimizer_settings = optimizer_settings,
symbol = network,
verbose = True,
)
data = load_cifar10_record(BATCH_SIZE)
info = solver.train(data)
identifier = 'cifar-residual-network-%d-%s-transition' % (N, mode)
pickle.dump(info, open('info/%s' % identifier, 'wb'))
optimizer_settings = {
'args': {
'momentum': 0.9
},
'initial_lr': lr,
'lr_scheduler': AtIterationScheduler(lr, lr_table),
'optimizer': 'SGD',
'weight_decay': 0.0001,
}
solver = MXSolver(
batch_size=BATCH_SIZE,
devices=(0, 1, 2, 3),
epochs=int(sys.argv[1]),
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
info = solver.train(data)
identifier = 'triple-state-transitory-residual-network'
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters, states = solver.export_parameters()
parameters = {
key: value
for key, value in parameters.items() if 'transition' in key
}