def train_model(real_model_folder):
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
batch_size = 100
# basic dataset holder
mnist = MNIST(device, "flat")
# creates the dataloader for pytorch
train_loader = DataLoader(dataset=mnist.train,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=mnist.validation,
batch_size=batch_size,
shuffle=False)
test_loader = DataLoader(dataset=mnist.test,
batch_size=batch_size,
shuffle=False)
real_model_param_path = os.path.join(model_path, real_model_folder,
"PiReal.param.pickle")
real_param = torch.load(real_model_param_path, map_location="cpu")
logit_net = PiLogitTanh(real_param, DiscreteWeights.TERNARY)
logit_net = logit_net.to(device)
logit_net.train_model(train_loader, validation_loader, test_loader, 200,
"MNIST-Pi-Tanh-Ternary", True)
def train_model():
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# basic dataset holder
mnist = MNIST(device, "2d")
# creates the dataloader for pytorch
batch_size = 100
train_loader = DataLoader(dataset=mnist.train,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=mnist.validation,
batch_size=batch_size,
shuffle=False)
test_loader = DataLoader(dataset=mnist.test,
batch_size=batch_size,
shuffle=False)
net = MnistAlternateDiscreteConv()
net = net.to(device)
num_epochs = 100
# will save metrics and model to disk. returns the path to metrics and saved model
return net.train_model(train_loader,
validation_loader,
test_loader,
num_epochs,
model_name="MNIST-conv-alt-conv")
def train_model(real_model_folder):
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
batch_size = 100
# basic dataset holder
mnist = MNIST(device, "2d")
# creates the dataloader for pytorch
train_loader = DataLoader(dataset=mnist.train,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=mnist.validation,
batch_size=batch_size,
shuffle=False)
test_loader = DataLoader(dataset=mnist.test,
batch_size=batch_size,
shuffle=False)
real_model_param_path = os.path.join(model_path, real_model_folder,
"ConvReal.param.pickle")
real_param = torch.load(real_model_param_path, map_location="cpu")
logit_net = ConvLogitSign(real_param, DiscreteWeights.TERNARY)
logit_net = logit_net.to(device)
# evaluate first logit model before training, train and evaluate again
logit_net.train_model(train_loader, validation_loader, test_loader, 200,
"MNIST-Conv-Ternary-Sign", True)
def train_model():
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# basic dataset holder
mnist = MNIST(device, "flat")
# creates the dataloader for pytorch
batch_size = 100
train_loader = DataLoader(dataset=mnist.train,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=mnist.validation,
batch_size=batch_size,
shuffle=False)
test_loader = DataLoader(dataset=mnist.test,
batch_size=batch_size,
shuffle=False)
net = PiReal()
net = net.to(device)
num_epochs = 100
# will save metrics and model to disk
return net.train_model(train_loader,
validation_loader,
test_loader,
num_epochs,
model_name="MNIST-pi-real")
def train_model():
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
# basic dataset holder
mnist = MNIST(device, "2d")
# creates the dataloader for pytorch
batch_size = 100
train_loader = DataLoader(dataset=mnist.train,
batch_size=batch_size,
shuffle=True)
validation_loader = DataLoader(dataset=mnist.validation,
batch_size=batch_size,
shuffle=False)
test_loader = DataLoader(dataset=mnist.test,
batch_size=batch_size,
shuffle=False)
net = ConvForcedQuantization()
net = net.to(device)
num_epochs = 200
net.train_model(train_loader, validation_loader, test_loader, num_epochs,
"MNIST-Conv-Forced-Quantization")