subset_dataset = get_subset_dataset(full_dataset=full_dataset,
examples_per_class=args.examples_per_class,
epc_seed=row.epc_seed,
root=osp.join(args.dataset_root, args.dataset),
train=True,
transform=transform,
download=True
)
loader = DataLoader(dataset=subset_dataset,
drop_last=False,
batch_size=args.batch_size)
C = row.num_classes
logger.info('Starting layerwise hessian decomposition...')
for layer_index, (layer_name, _) in enumerate(model.named_parameters()):
logger.info('layer_index: {}\nlayer_name: {}'.format(layer_index, layer_name))
logger.info('decomposing hessian')
Hess = LayerHessian(crit='CrossEntropyLoss',
loader=loader,
device=device,
model=model,
num_classes=C,
layer_name=layer_name,
hessian_type='Hessian',
init_poly_deg=64,
poly_deg=128,
spectrum_margin=0.05,
poly_points=1024,
SSI_iters=128
)
train=True,
transform=transform,
download=True
)
loader = DataLoader(dataset=subset_dataset,
drop_last=False,
batch_size=args.batch_size)
C = row.num_classes
logger.info('Starting layerwise hessian decomposition...')
ub = list()
lb = list()
for layer_name, param in model.named_parameters():
if 'weight' not in layer_name:
continue
if '.bn' in layer_name:
continue
logger.info('layer_name: {}'.format(layer_name))
Hess = LayerHessian(crit='CrossEntropyLoss',
loader=loader,
device=device,
model=model,
num_classes=C,
layer_name=layer_name,
hessian_type='Hessian',
init_poly_deg=64,
poly_deg=128,
spectrum_margin=0.05,
model_weights_path, results_path = download_model(model_url, ckpt_dir)
df = pd.read_csv(results_path)
row = df.iloc[0]
if args.model in [
'VGG11_bn', 'Resnet18', 'DenseNet3_40', 'LeNet', 'MobileNet'
]:
model = Network().construct(args.model, row)
else:
raise Exception('Unknown model argument: {}'.format(args.model))
if args.print_layers:
total_params = sum(p.numel() for p in model.parameters()
if p.requires_grad)
print('Total number of parameters: ', total_params, '\nlayers:')
pp.pprint(list(name for (name, _) in model.named_parameters()))
exit(code=0)
state_dict = torch.load(model_weights_path,
map_location=lambda storage, loc: storage)
if args.new:
state_dict = state_dict['model']
model.load_state_dict(state_dict, strict=True)
model = model.to(device)
model = model.eval()
mean, std = get_mean_std(args.dataset)
pad = int((row.padded_im_size - row.im_size) / 2)
transform = transforms.Compose([
transforms.Pad(pad),
model_weights_path, results_path = download_model(model_url, ckpt_dir)
df = pd.read_csv(results_path)
row = df.iloc[0]
if args.model in [
'VGG11_bn', 'ResNet18', 'DenseNet3_40', 'MobileNet', 'LeNet'
]:
model = Network().construct(args.model, row)
else:
raise Exception('Unknown model argument: {}'.format(args.model))
if args.print_layers:
total_params = sum(p.numel() for p in model.parameters()
if p.requires_grad)
print('Total number of parameters: ', total_params, '\nlayers:')
pp.pprint(list(name for (name, _) in model.named_parameters()))
exit(code=0)
state_dict = torch.load(model_weights_path,
map_location=lambda storage, loc: storage)
model.load_state_dict(state_dict, strict=True)
model = model.to(device)
if args.cuda is None:
gpus = torch.cuda.device_count()
if gpus > 1:
model = nn.DataParallel(model, device_ids=range(gpus))
model = model.eval()
mean, std = get_mean_std(args.dataset)
pad = int((row.padded_im_size - row.im_size) / 2)
transform = transforms.Compose([
