def bruna10(pretrained_path=False, hidden_size=10, **kwargs):
"""Constructs a very simple convNet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = singleHiddenFullyConnected(layers=[3072, hidden_size, 100])
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def conv1020tanh(pretrained_path=False, **kwargs):
"""Constructs a very simple convNet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = conv22tanh(layers=[3, 10, 32, 400, 100], im_dim=32)
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def alex6464(pretrained_path=False, **kwargs):
"""alex like convnet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = AlexNet(layers=[3, 64, 128, 512, 128, 100])
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def alex6464raw(pretrained_path=False, **kwargs):
"""alex like convnet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = AlexNet(raw=True)
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def conv6464relu(pretrained_path=False, **kwargs):
"""alex like convnet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = convAlexrelu(layers=[3, 64, 64, 192, 10], k_dims=[5, 5], im_dim=32)
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def conv1020relu(pretrained_path=False, **kwargs):
"""todo
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = conv22relu(layers=[3, 10, 20, 100, 10], im_dim=32)
if pretrained_path:
return loadNet(pretrained_path, model)
return model
def convTest(pretrained_path=False, **kwargs):
"""Constructs a very simple convNet
Args:
pretrained_path (bool): If True, returns the pretrained model on the path
"""
model = conv22relu(layers=[1, 2, 4, 32, 10])
if pretrained_path:
return loadNet(pretrained_path, model)
return model
elif args.dataset=='cifar10':
import models.cifar10 as my_models
model = getattr(my_models, args.model)
elif args.dataset=='mnist':
import models.mnist as my_models
model = getattr(my_models, args.model)
else:
print("Unknown dataset: ",args.dataset);
sys.exit()
#Load network
if args.load == 'nil':
model = model(hidden_size=args.hidden_size)
iniPeriod=0
else:
model=loadNet(args.load,model)
from re import search
iniPeriod=1+int(search('_',args.model,'_(.+?).pyT',args.load).group(1))
#################################
# Initialization of the weights #
#################################
# Function that yields all the optimizable parameters of the network (the network's size)
def getNumParam(mymodel):
tot=0
for item in list(mymodel.parameters()):
n_layer=item.numel()
tot+=n_layer
return tot
