},
'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)
import cPickle as pickle
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters = solver.export_parameters()
pickle.dump(parameters, open('parameters/%s' % identifier, 'wb'))
'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
}
states = {key: value for key, value in states.items() if 'transition' in key}
pickle.dump((parameters, states), open('parameters/%s' % identifier, 'wb'))