},
'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'))
if not ('data' in name or 'label' in name)
}
state_names = loss_group.list_auxiliary_states()
states = {
name: mx.nd.zeros(shape, context)
for name, shape in zip(state_names, state_shapes)
}
for name, array in states.items():
initializer(name, array)
executor = loss_group.bind(context, arguments, gradients, aux_states=states)
from data_utilities import load_mnist
original = load_mnist(path='stretched_canvas_mnist',
scale=1,
shape=(1, 56, 56))
stretched = load_mnist(path='stretched_mnist', scale=1, shape=(1, 56, 56))
from mxnet.io import NDArrayIter as Iterator
iterator = Iterator(stretched[0], stretched[1], args.batch_size, shuffle=True)
unpack = lambda batch: (batch.data[0], batch.label[0])
n_iterations = 0
for batch in iterator:
data, labels = unpack(batch)
arguments['data'][:] = data
arguments['labels'][:] = labels
executor.forward(is_train=True)
executor.backward()
optimizer_settings = {
'args': {
'momentum': 0.9
},
'initial_lr': 0.1,
'optimizer': 'SGD'
}
solver = MXSolver(
batch_size=64,
devices=(configs.gpu_index, ),
epochs=50,
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
training_data, training_labels, _, _, _, _ = load_mnist(shape=(1, 28, 28))
_, _, validation_data, validation_labels, test_data, test_labels = load_mnist(
path='shrinked_mnist', shape=(1, 28, 28))
data = training_data, training_labels, validation_data, validation_labels, test_data, test_labels
info = solver.train(data)
identifier = 'shrinked-mnist-fixed-attention-network'
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters = solver.export_parameters()
pickle.dump(parameters, open('parameters/%s' % identifier, 'wb'))
BATCH_SIZE = 128
lr = 0.1
lr_table = {10000 : 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 = 50,
initializer = PReLUInitializer(),
optimizer_settings = optimizer_settings,
symbol = network,
verbose = True,
)
data = load_mnist(path='rescaled_mnist', shape=(1, 42, 42))
info = solver.train(data)
identifier = 'rescaled-mnist-baseline-network-%d-%s' % (N, sys.argv[2])
pickle.dump(info, open('info/%s' % identifier, 'wb'))
parameters = solver.export_parameters()
pickle.dump(parameters, open('parameters/%s' % identifier, 'wb'))
parser.add_argument('--n_filters', type=int, default=4)
parser.add_argument('--n_layers', type=int, default=3)
parser.add_argument('--n_scales', type=int, default=3)
parser.add_argument('--n_units', type=int, default=16)
args = parser.parse_args()
import mxnet as mx
from mxnet.context import Context
context = mx.cpu() if args.gpu_index < 0 else mx.gpu(args.gpu_index)
Context.default_ctx = context
unpack_batch = lambda batch : \
(batch.data[0].as_in_context(context), batch.label[0].as_in_context(context))
from data_utilities import load_mnist
data = load_mnist(path=args.path, normalize=True, shape=(1, 112, 112))
# data = load_mnist(path=args.path, normalize=True, shape=(1, 56, 56))
from mxnet.io import NDArrayIter
training_data = NDArrayIter(data[0], data[1], batch_size=args.batch_size)
validation_data = NDArrayIter(data[2], data[3], batch_size=args.batch_size)
test_data = NDArrayIter(data[4], data[5], batch_size=args.batch_size)
model = MSPCNN(args.n_layers, args.n_filters, args.n_scales, args.n_units)
updater = Updater(model, update_rule='adam', lr=args.lr)
# updater = Updater(model, update_rule='sgd_momentum', lr=1e-1, momentum=0.9)
import numpy as np
from mxnet.contrib.autograd import compute_gradient
import minpy.nn.utils as utils
parser.add_argument('--n_filters', type=int, default=4)
parser.add_argument('--n_layers', type=int, default=3)
parser.add_argument('--n_scales', type=int, default=3)
parser.add_argument('--n_units', type=int, default=16)
args = parser.parse_args()
import mxnet as mx
from mxnet.context import Context
context = mx.cpu() if args.gpu_index < 0 else mx.gpu(args.gpu_index)
Context.default_ctx = context
unpack_batch = lambda batch : \
(batch.data[0].as_in_context(context), batch.label[0].as_in_context(context))
from data_utilities import load_mnist
data = load_mnist(path=args.path, normalize=True, shape=(1, 56, 56))
# data = load_mnist(path=args.path, normalize=True, shape=(1, 112, 112))
from mxnet.io import NDArrayIter
training_data = NDArrayIter(data[0], data[1], batch_size=args.batch_size)
validation_data = NDArrayIter(data[2], data[3], batch_size=args.batch_size)
test_data = NDArrayIter(data[4], data[5], batch_size=args.batch_size)
model = ReferentialCNN(args.n_layers, args.n_filters, args.n_scales,
args.n_units)
updater = Updater(model, update_rule='adam', lr=args.lr)
# updater = Updater(model, update_rule='sgd_momentum', lr=1e-1, momentum=0.9)
import numpy as np
from mxnet.contrib.autograd import compute_gradient
import minpy.nn.utils as utils
optimizer_settings = {
'args': {
'momentum': 0.9
},
'initial_lr': 0.1,
'optimizer': 'SGD'
}
solver = MXSolver(
batch_size=64,
devices=(configs.gpu_index, ),
epochs=30,
initializer=PReLUInitializer(),
optimizer_settings=optimizer_settings,
symbol=network,
verbose=True,
)
data = []
data.extend(load_mnist(path='stretched_mnist', scale=1, shape=(1, 56, 56))[:2])
data.extend(
load_mnist(path='stretched_canvas_mnist', scale=1, shape=(1, 56, 56))[2:])
info = solver.train(data)
identifier = 'shrinked-mnist-plain-network-%d-%s' % (configs.n_plain_layers,
configs.postfix)
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': 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(scale=1, shape=(1, 28, 28))
info = solver.train(data)
postfix = '-' + args.postfix if args.postfix else ''
identifier = 'residual-network-on-standard-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'))