# download data if necessary
def _download(data_dir):
if not os.path.isdir(data_dir):
os.system('mkdir ' + data_dir)
os.chdir(data_dir)
if (not os.path.exists('train.rec')) or \
(not os.path.exists('test.rec')):
os.system('wget http://webdocs.cs.ualberta.ca/~bx3/data/cifar10.zip')
os.system('unzip -u cifar10.zip')
os.system('mv cifar/* .; rm -rf cifar; rm cifar10.zip')
os.chdir('..')
# network
from symbol_resnet import get_symbol
net = get_symbol(num_class=10)
# data
def get_iterator(args, kv):
kargs = dict(
data_shape=(3, 32, 32),
# Use mean and scale works equally well
# We use BatchNorm after data for simplicity
# mean_r=127,
# mean_g=127,
# mean_b=127,
# scale=1 / 60
)
if '://' not in args.data_dir:
_download(args.data_dir)
from symbol_resnet import get_symbol
def check_speed(sym, ctx, scale=1.0, N=100):
exe = sym.simple_bind(grad_req='write', **ctx)
init = [
np.random.normal(size=arr.shape, scale=scale) for arr in exe.arg_arrays
]
for arr, iarr in zip(exe.arg_arrays, init):
arr[:] = iarr.astype(arr.dtype)
exe.forward(is_train=False)
#exe.backward(exe.outputs[0])
exe.outputs[0].wait_to_read()
tic = time.time()
for i in range(N):
exe.forward(is_train=False)
exe.backward(exe.outputs[0])
exe.outputs[0].wait_to_read()
return (time.time() - tic) * 1000.0 / N
sym_res = get_symbol()
print(mx.visualization.print_summary(sym_res, shape={'data':
(1, 3, 224, 224)}))
#ctx_list = [{'ctx': mx.gpu(0), 'data': (1, 3, 224, 224)}]
#for ctx in ctx_list:
# print(ctx, check_speed(sym_res, ctx, N=10))
parser.add_argument('--num-epochs',
type=int,
default=200,
help='the number of training epochs')
parser.add_argument('--load-epoch',
type=int,
help='load the model on an epoch using the model-prefix')
parser.add_argument('--kv-store',
type=str,
default='local',
help='the kvstore type')
args = parser.parse_args()
# network
from symbol_resnet import get_symbol
net = get_symbol(num_class=10)
# data
def get_iterator(args, kv):
kargs = dict(
data_shape=(3, 32, 32),
# Use mean and scale works equally well
# We use BatchNorm after data for simplicity
# mean_r=127,
# mean_g=127,
# mean_b=127,
# scale=1 / 60
)
train = mx.io.ImageRecordIter(
type=int,
default=4,
help='the number of training epochs')
parser.add_argument('--load-epoch',
type=int,
help='load the model on an epoch using the model-prefix')
parser.add_argument('--kv-store',
type=str,
default='local',
help='the kvstore type')
args = parser.parse_args()
# network
from symbol_resnet import get_symbol
net = get_symbol(num_classes=1000,
num_layers=50,
image_shape='3, 224, 224',
conv_workspace=512)
# data
def get_iterator(args, kv):
kargs = dict(
data_shape=(3, 224, 224),
# Use mean and scale works equally well
# We use BatchNorm after data for simplicity
# mean_r=127,
# mean_g=127,
# mean_b=127,
# scale=1 / 60
)