def main(config):
acc_dict = {}
trainset, valset = CIFAR100('data/')
train_dict, val_dict = continual_wrapper(trainset, valset, num_tasks=10)
for i in range(3, 10):
del train_dict[i]
del val_dict[i]
net = seresnet20_cifar100().to(config.device)
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(net.parameters(),
lr=config.lr,
momentum=config.mo)
for task in range(config.num_tasks):
trainloader = torch.utils.data.DataLoader(train_dict[task],
batch_size=config.bs)
valloader = torch.utils.data.DataLoader(val_dict[task],
batch_size=config.bs)
learn(net, trainloader, valloader, criterion, optimizer, config)
acc_dict[task] = {}
for i in range(task + 1):
valloader = torch.utils.data.DataLoader(val_dict[i],
batch_size=config.bs)
acc_dict[task][i] = test(net, valloader, criterion, config)[1]
visualise(acc_dict)
if __name__ == '__main__':
board = board_init(6, 6)
if sys.argv[1] != "0":
file_name = sys.argv[1]
else:
env_parameter = [6, 6, 0, 33]
agent_parameter = [5, 0.2, 1.0, 0.9,
0.0] #numActions, epsilon, gamma, alpha, init value
task_env, task_agent = experiment.RL_init(env_parameter,
agent_parameter)
task_agent.behavior_value = np.load("B_value.npy")
task_agent.Q_value = np.load("Q.npy")
experiment.test(task_env, task_agent, 200)
file_name = "track.csv"
with open(file_name, 'rb') as f:
reader = csv.reader(f)
track = list(reader)
print "Start..."
i = 0
count = 0
for item in track:
current_board = copy.deepcopy(board)
i += 1
x, y = utils.to_cord(int(item[0]), 6)
current_board[5 - y][x] = '@'
if not os.path.exists('results/0_results.p'):
print "No results found! Please run some setupExperiments.py before viewing results!"
quit()
with open('results/0_results.p') as f:
x = pickle.load(f)
params = [y[0] for y in x[0]['params']]
tab = PrettyTable(["Run", "NumEvals", "ValObj"] + params + ["TestObj"])
tab.align = 'l'
print "Generating Results Table/Testing Hyperparameters..."
lb = LoadBar(np.min([numSplits, 30]))
lb.setup()
for i in range(numSplits):
if lb.test(i, numSplits): lb += 1
with open("results/" + str(i) + "_results.p") as f:
x = pickle.load(f)
objectives = np.array([a['objective'] for a in x])
cMin = x[np.argmin(objectives)]
paramDict = dict(cMin['params'])
sys.path.append(str(i))
from experiment import test
acc = test('experiments/' + str(i) + "/", paramDict)
tab.add_row([i, len(objectives), "{:.4f}".format(cMin['objective'])]
+ ["{:.4f}".format(paramDict[p][0]) for p in params]
+ ["{:.4f}".format(acc)])
lb.clear()
print tab
def testTestFGSMSVHN(self):
test(num_gpus=2,
total_batch_size=200,
image_size=28,
summary_dir='debug/summary/caps/svhn/Default/',
load_test_path='debug/data/caps/svhn/test_FGSM.npz')
def testTestMNIST(self):
test(num_gpus=2,
total_batch_size=200,
image_size=28,
summary_dir='debug/summary/caps/mnist/Default/',
load_test_path='debug/data/caps/mnist/test.npz')
print('\n{:-^50}'.format(' Network Initialized '))
utils.print_network(model)
print('-' * 50 + '\n')
if args.stage == 'train' or args.stage == 'both':
print(opts)
experiment.train(opts,
model,
training_dataset,
validation_dataset,
opts.train_epochs,
resume_from_epoch=resume_from_epoch)
if args.stage == 'test' or args.stage == 'both':
print('\n{:-^50}'.format(' Testing Model '))
experiment.test(opts, model, testing_dataset, save_loss=True)
opts.jobs += 1
if opts.hyperparameter == False:
# Selected hyperparameters
param_grid = {
'lr': 1e-4,
'batch_size': 4,
'loss_type': 'l1',
'loss_formulation': 'absolute',
'image_normalization': False,
'compensated_target': False,
}
secondsPerSplit = params['secondsPerSplit']
if not os.path.exists('results/0_results.p'):
print "No results found! Please run some setupExperiments.py before viewing results!"
quit()
with open('results/0_results.p') as f:
x = pickle.load(f)
params = [y[0] for y in x[0]['params']]
tab = PrettyTable(["Run", "NumEvals", "ValObj"] + params + ["TestObj"])
tab.align = 'l'
print "Generating Results Table/Testing Hyperparameters..."
lb = LoadBar(np.min([numSplits, 30]))
lb.setup()
for i in range(numSplits):
if lb.test(i, numSplits): lb += 1
with open("results/" + str(i) + "_results.p") as f:
x = pickle.load(f)
objectives = np.array([a['objective'] for a in x])
cMin = x[np.argmin(objectives)]
paramDict = dict(cMin['params'])
sys.path.append(str(i))
from experiment import test
acc = test('experiments/' + str(i) + "/", paramDict)
tab.add_row([i, len(objectives), "{:.4f}".format(cMin['objective'])] +
["{:.4f}".format(paramDict[p][0])
for p in params] + ["{:.4f}".format(acc)])
lb.clear()
print tab