def get_model(self):
# get symbol from symbol.py
self.symbol = symbol.get_symbol(self, self.label_num, self.ignore_label, self.aspp, self.aspp_stride,
self.atrous_type, self.bn_use_global_stats, self.relu_type)
# load model
if self.load_model_prefix is not None and self.load_epoch > 0:
self.symbol, self.arg_params, self.aux_params = \
mx.model.load_checkpoint(os.path.join(self.load_model_dir, self.load_model_prefix), self.load_epoch)
parser.add_argument('--num-examples', type=int, default=int(2140 * 0.8),
help='the number of training examples')
# need log to file?
parser.add_argument('--log-dir', type=str, default="/tmp/",
help='directory of the log file')
parser.add_argument('--load-epoch', type=int,
help="load the model on an epoch using the model-prefix")
parser.add_argument('--save-model-prefix', type=str,
help='the prefix of the model to save')
# todo statistic about mean data
args = parser.parse_args()
import symbol
net = symbol.get_symbol(output_dim = 30)
from data import FileIter
train = FileIter(
eval_ratio = 0.2,
is_val = False,
data_name = "data",
batch_size = args.batch_size,
label_name = "lr_label"
)
val = FileIter(
eval_ratio = 0.2,
is_val = True,
data_name = "data",
import mxnet as mx
import dataLoader, deploy, symbol, debug
network = symbol.get_symbol(num_classes = 10)
# print network.list_arguments()
net = mx.mod.Module(symbol = network, context = mx.gpu(0), fixed_param_names = [ "fc1_weight", "fc1_bias", "fc2_weight", "fc2_bias"])
trainDataIter, valDataIter = dataLoader.get_data_iter()
num_epoch = 120
try:
net.fit(train_data = trainDataIter,
eval_data = valDataIter,
epoch_end_callback = debug.epoch_end_callback,
eval_end_callback = debug.eval_end_callback,
eval_metric = deploy.get_eval_metric(),
optimizer = "sgd",
initializer = deploy.get_initializer(),
num_epoch = num_epoch,
begin_epoch = deploy.get_begin_epoch()
)
results = net.score(valDataIter, deploy.get_eval_metric(), reset = True)
print "[Info] Validation Result:", results
results = net.score(trainDataIter, deploy.get_eval_metric(), reset = True)
print "[Info] Training Result:", results
print "[Info] Saving Parameters..."
net.save_params('models/test2_{}.params'.format(str(num_epoch)))
def actualize(self, ev, env):
"""
Return the constant value. The dispatched event is simply ignored.
"""
ll = [s for s in symbol.get_symbol(self.name)]
return ll
def __function_name(self):
s = symbol.get_symbol(self.event.function)
if s: return s[0]
else: return None
def actualize(self, ev, env):
"""
Return the constant value. The dispatched event is simply ignored.
"""
ll = [s for s in symbol.get_symbol(self.name)]
return ll
def resolve(s):
r = symbol.get_symbol(s)
return set(r)
