def main(batchsize=10, lr=0.01, epochs=10, frac_dset=1., fname=''):
tform = transforms.transformation(
[transforms.ToTensor(),
transforms.Standardize()])
trainset = mnist_dataset(train=True,
download=True,
frac=frac_dset,
transform=tform)
trainloader = dataloader(trainset, batchsize)
testset = mnist_dataset(train=False, download=True, transform=tform)
testloader = dataloader(testset, batchsize)
model = Model()
optimizer = optim.SGD(model.get_parameters(), lr=lr)
criterion = loss.MSELoss()
if fname == '':
fout = None
else:
fout = open(fname, 'w')
t1 = time.time()
for epoch in range(epochs):
train(model, criterion, optimizer, trainloader, epoch, fout=fout)
test(model, criterion, testloader, epoch, fout=fout)
if epoch % 10 == 9:
optimizer._lr *= 0.1
t2 = time.time()
output_str = 'Total Training Time: %.2f seconds' % (t2 - t1)
if fout is None:
print(output_str)
else:
fout.write(output_str + '\n')
fout.close()
def main(batchsize=10, lr=0.01, epochs=10, k=5, frac_dset=1., fname=''):
tform = transforms.transformation([transforms.ToTensor()])
trainset = mnist_dataset(train=True, download=True, frac=frac_dset, oneHot=False, transform=tform)
trainset._images = trainset._images / 255.
mean = np.expand_dims(np.mean(trainset._images, axis=0), 0)
std = np.expand_dims(np.std(trainset._images, axis=0), 0)
std[std <= 0.03] = 1
trainset._images = (trainset._images - mean) / std
trainloader = dataloader(trainset, batchsize)
testset = mnist_dataset(train=False, download=True, oneHot=False, transform=tform)
testset._images = testset._images / 255.
testset._images = (testset._images - mean) / std
testloader = dataloader(testset, batchsize)
model = Model(k)
optimizer = optim.SGD(model.get_parameters(), lr=lr)
criterion = loss.MSELoss()
if fname == '':
fout = None
else:
fout = open(fname, 'w')
t1 = time.time()
for epoch in range(epochs):
train(model, criterion, optimizer, trainloader, epoch, fout=fout)
test(model, criterion, testloader, epoch, fout=fout)
t2 = time.time()
output_str = 'Total Training Time: %.2f seconds' % (t2-t1)
if fout is None:
print(output_str)
else:
fout.write(output_str+'\n')
fout.close()