def __init__(self, configuration):
base_configuration = {
'kernel_type':
DoGKernel,
'kernels_conf': [[3, 3 / 9, 6 / 9], [3, 6 / 9, 3 / 9],
[7, 7 / 9, 14 / 9], [7, 14 / 9, 7 / 9],
[13, 13 / 9, 26 / 9], [13, 26 / 9, 13 / 9]],
'padding':
6,
'threshold':
50,
'use_threshold':
True
}
self.configuration = u.populate_configuration(configuration,
base_configuration)
# creates kernels and pad smaller kernels to fit biggest one
self.max_kernel_size = max(
[conf[0] for conf in self.configuration['kernels_conf']])
kernels = []
for conf in self.configuration['kernels_conf']:
kernel = self.configuration['kernel_type'](*conf)().unsqueeze(0)
pad = (self.max_kernel_size - conf[0]) // 2
kernels.append(
torch.nn.functional.pad(kernel, (pad, pad, pad, pad)))
self.kernels = torch.stack(kernels)
def __init__(self, configuration={}):
self.base_config = {
'layers_config': [{
'in_channels': 2,
'out_channels': 32,
'kernel_size': 5,
'weight_mean': 0.8,
'weight_std': 0.05
}, {
'in_channels': 32,
'out_channels': 150,
'kernel_size': 2,
'weight_mean': 0.8,
'weight_std': 0.05
}],
'learning_rates': (0.004, -0.003),
'max_ap':
torch.tensor([0.15]),
'an_update':
-0.75,
'thresholds': [10, 1],
'timestep_update_lr':
500,
'pooling_vars': [{
'kernel_size': 2,
'stride': 2,
'padding': 1
}] * 2,
'n_winners': [5, 8],
'inhibition_radius': [2, 1],
'batch_size':
1,
'dataset_path':
'data/',
'n_epochs': [2, 20],
'save_path':
'model/sdcnne_exp.pt'
}
self.config = u.populate_configuration(configuration, self.base_config)
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
self.config['max_ap'] = self.config['max_ap'].to(self.device)
pad_fn = lambda x: (x - 1) // 2 + ((x - 1) % 2 > 0)
self.paddings_train = [[pad_fn(conf['kernel_size'])] * 4
for conf in self.config['layers_config']]
self.paddings_forward = [
pad_fn(conf['kernel_size'])
for conf in self.config['layers_config']
]
self.set_metadata()
self.set_dataloader()
self.instanciate_model()
def __init__(self, configuration={}):
base_config = {
'kernel_type': toci.DoGKernel,
'kernels_conf': [[7, 1, 2], [7, 2, 1]],
'padding': 3,
'threshold': 50,
'use_threshold': True,
'normalization_radius': 8,
'n_time_steps': 15
}
self.config = u.populate_configuration(base_config, configuration)
self.to_tensor = torchvision.transforms.ToTensor()
self.filters = toci.Filter(self.config)
argparser.add_argument('--reduce_dim', default=False, type=ast.literal_eval)
argparser.add_argument('--lr', default=1e-4, type=float)
argparser.add_argument('--smoothing_epsilon', default=0.1, type=float)
argparser.add_argument('--save_path', default='convnet/', type=str)
argparser.add_argument('--plot_attention', default=False, type=ast.literal_eval)
argparser.add_argument('--eval_step', default=10, type=int)
argparser.add_argument('--max_epochs', default=500, type=int)
argparser.add_argument('--train_acc_step', default=5, type=int)
argparser.add_argument('--load_model', default=False, type=ast.literal_eval)
argparser.add_argument('--clip_grad', default=False, type=ast.literal_eval)
argparser.add_argument('--weight_decay', default=0., type=float) # L2 regularization -> 0.01
argparser.add_argument('--l1_reg', default=0., type=float) # L1 regularization -> 0.001
argparser.add_argument('--warmup', default=100, type=int)
argparser.add_argument('--clip_grad_val', default=0.1, type=float)
argparser.add_argument('--logfile', default='_convnet_experiments_feedback_logs.txt', type=str)
args = argparser.parse_args()
# logging.basicConfig(stream=sys.stdout, level=logging.INFO)
logging.basicConfig(filename=args.logfile, filemode='a', level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
settings = u.populate_configuration(settings, vars(args))
global plotter
plotter = u.VisdomPlotter(env_name='ConvNet Experiments')
if not os.path.isdir(settings['save_path']):
os.makedirs(settings['save_path'])
launch_experiment(settings)