if torch.cuda.is_available():
net = net.cuda()
for name, param in net.named_parameters():
param.requires_grad = False
if name[5] == str(
hyper_params['stage']) and hyper_params['stage'] != 0:
param.requires_grad = True
elif (name[0] == 'b'
or name[0] == 'c') and hyper_params['stage'] == 0:
param.requires_grad = True
# saving outputs of all Basic Blocks
mdl = learn.model
sf = [
SaveFeatures(m)
for m in [mdl[0][2], mdl[0][4], mdl[0][5], mdl[0][6], mdl[0][7]]
]
sf2 = [
SaveFeatures(m) for m in
[net.relu2, net.layer1, net.layer2, net.layer3, net.layer4]
]
project_name = 'tl-kd-' + hyper_params['model'] + '-' + hyper_params[
'dataset']
experiment = Experiment(api_key="IOZ5docSriEdGRdQmdXQn9kpu",
project_name=project_name,
workspace="akshaykvnit")
experiment.log_parameters(hyper_params)
if hyper_params['stage'] == 0:
filename = '../saved_models/' + str(
if torch.cuda.is_available():
net = net.cuda()
# Freezing all parameters except of current stage
for name, param in net.named_parameters():
param.requires_grad = False
if name[5] == str(
hyper_params['stage']) and hyper_params['stage'] != 0:
param.requires_grad = True
elif (name[0] == 'b'
or name[0] == 'c') and hyper_params['stage'] == 0:
param.requires_grad = True
# saving outputs of all Basic Blocks
sf = [
SaveFeatures(m)
for m in [model.relu2, model.layer1, model.layer2, model.layer3]
]
sf2 = [
SaveFeatures(m)
for m in [net.relu2, net.layer1, net.layer2, net.layer3]
]
experiment = Experiment(api_key="IOZ5docSriEdGRdQmdXQn9kpu",
project_name="kd6",
workspace="akshaykvnit")
experiment.log_parameters(hyper_params)
if hyper_params['stage'] == 0:
filename = '../saved_models/' + str(
hyper_params['dataset']) + '/' + str(
hyper_params['student']) + '_stage' + str(
filename = '../saved_models/' + str(hyper_params['dataset']) + '/less_data' + str(hyper_params['perc']) + '/' + hyper_params['model'] + '_stage' + str(hyper_params['stage']) + '/model' + str(hyper_params['repeated']) + '.pt'
net.load_state_dict(torch.load(filename, map_location = 'cpu'))
if torch.cuda.is_available() :
net = net.cuda()
for name, param in net.named_parameters() :
param.requires_grad = False
if name[5] == str(hyper_params['stage']) and hyper_params['stage'] != 0 :
param.requires_grad = True
elif (name[0] == 'b' or name[0] == 'c') and hyper_params['stage'] == 0 :
param.requires_grad = True
# saving outputs of all Basic Blocks
mdl = learn.model
sf = [SaveFeatures(m) for m in [mdl[0][2], mdl[0][4], mdl[0][5], mdl[0][6], mdl[0][7]]]
sf2 = [SaveFeatures(m) for m in [net.relu2, net.layer1, net.layer2, net.layer3, net.layer4]]
project_name = 'kd-ld' + str(hyper_params['perc']) + '-' + hyper_params['dataset'] + '-' + hyper_params['model']
experiment = Experiment(api_key="1jNZ1sunRoAoI2TyremCNnYLO", project_name = project_name, workspace = "akshaykvnit")
experiment.log_parameters(hyper_params)
if hyper_params['stage'] == 0 :
os.makedirs('../saved_models/' + str(hyper_params['dataset']) + '/less_data' + str(hyper_params['perc']) + '/' + hyper_params['model'] + '_stage' + str(hyper_params['stage']), exist_ok = True)
filename = '../saved_models/' + str(hyper_params['dataset']) + '/less_data' + str(hyper_params['perc']) + '/' + hyper_params['model'] + '_stage' + str(hyper_params['stage']) + '/model' + str(hyper_params['repeated']) + '.pt'
else :
os.makedirs('../saved_models/' + str(hyper_params['dataset']) + '/less_data' + str(hyper_params['perc']) + '/' + hyper_params['model'] + '_stage' + str(hyper_params['stage'] + 1), exist_ok = True)
filename = '../saved_models/' + str(hyper_params['dataset']) + '/less_data' + str(hyper_params['perc']) + '/' + hyper_params['model'] + '_stage' + str(hyper_params['stage'] + 1) + '/model' + str(hyper_params['repeated']) + '.pt'
optimizer = torch.optim.Adam(net.parameters(), lr = hyper_params["learning_rate"])
total_step = len(data.train_ds) // hyper_params["batch_size"]
if torch.cuda.is_available():
net = net.cuda()
for name, param in net.named_parameters():
param.requires_grad = False
if name[0] == str(hyper_params['stage'] +
1) and hyper_params['stage'] != 0:
param.requires_grad = True
elif name[0] == str(
hyper_params['stage']) and hyper_params['stage'] == 0:
param.requires_grad = True
# saving outputs of all Basic Blocks
mdl = learn.model
sf = [
SaveFeatures(m)
for m in [mdl[0][2], mdl[0][4], mdl[0][5], mdl[0][6]]
]
sf2 = [SaveFeatures(m) for m in [net[0], net[2], net[3], net[4]]]
experiment = Experiment(api_key="IOZ5docSriEdGRdQmdXQn9kpu",
project_name="kd5",
workspace="akshaykvnit")
experiment.log_parameters(hyper_params)
if hyper_params['stage'] == 0:
filename = '../saved_models/' + str(
hyper_params['dataset']) + '/small_stage' + str(
hyper_params['stage']) + '/model' + str(
hyper_params['repeated']) + '.pt'
else:
filename = '../saved_models/' + str(
elif hyper_params['model'] == 'resnet14':
net = resnet14(pretrained=False, progress=False)
elif hyper_params['model'] == 'resnet18':
net = resnet18(pretrained=False, progress=False)
elif hyper_params['model'] == 'resnet20':
net = resnet20(pretrained=False, progress=False)
elif hyper_params['model'] == 'resnet26':
net = resnet26(pretrained=False, progress=False)
if torch.cuda.is_available():
net = net.cuda()
# saving outputs of all Basic Blocks
mdl = learn.model
# for all 5 feature maps
sf = [SaveFeatures(m) for m in [mdl[0][5]]]
sf2 = [SaveFeatures(m) for m in [net.layer2]]
num_fm = 1
project_name = 'trad-kd-ld-' + hyper_params['model'] + '-' + hyper_params[
'dataset']
experiment = Experiment(api_key="1jNZ1sunRoAoI2TyremCNnYLO",
project_name=project_name,
workspace="akshaykvnit")
experiment.log_parameters(hyper_params)
optimizer = torch.optim.Adam(net.parameters(),
lr=hyper_params["learning_rate"])
total_step = len(data.train_ds) // hyper_params["batch_size"]
train_loss_list = list()
val_loss_list = list()
min_val = 0