def __init__(self, dataset_name, alpha=1, beta=0.6, gamma=1, sigma=0.001, l2=1e-5, \
latent_dim=10, hidden_dim=100, num_layers=3,epochs=100, batch_size=32,\
dropout=0.2, batch_norm=True, l_rate=1e-05, restore=False, plot=False):
args=dict()
args['model_type']=0
args['model_name']='AE'
args['dataset_name']=dataset_name
args['alpha']=alpha
args['beta']=beta
args['gamma']=gamma
args['sigma']=sigma
args['l2']=l2
args['latent_dim']=latent_dim
args['hidden_dim']=hidden_dim
args['num_layers']=num_layers
args['epochs']=epochs
args['batch_size']=batch_size
args['dropout']=dropout
args['batch_norm']=batch_norm
args['l_rate']=l_rate
args['train']=True if restore==False else False
args['plot']=plot
args['restore']=restore
args['early_stopping']=1
dirs = ['checkpoint_dir','summary_dir', 'result_dir', 'log_dir' ]
for d in dirs:
args[d] = d
self.config = utils.Config(args)
def setup_logging(self):
experiments_root_dir = 'experiments'
self.config.model_name = get_model_name_CapEM(self.config.model_name,
self.config)
self.config.summary_dir = os.path.join(
experiments_root_dir + "\\" + self.config.log_dir + "\\",
self.config.model_name)
self.config.checkpoint_dir = os.path.join(
experiments_root_dir + "\\" + self.config.checkpoint_dir + "\\",
self.config.model_name)
self.config.results_dir = os.path.join(
experiments_root_dir + "\\" + self.config.result_dir + "\\",
self.config.model_name)
#Flags
flags_list = ['train', 'restore', 'results', 'plot', 'early_stopping']
self.flags = utils.Config({
your_key: self.config.__dict__[your_key]
for your_key in flags_list
})
# create the experiments dirs
utils.create_dirs([
self.config.summary_dir, self.config.checkpoint_dir,
self.config.results_dir
])
utils.save_args(self.config.__dict__, self.config.summary_dir)
def __init__(self,
dataset_name,
cell_size=8.0,
hidden_caps=4,
num_layers=3,
epochs=100,
batch_size=32,
l_rate=1e-05,
restore=0):
args = dict()
args['model_type'] = 0
args['model_name'] = 'CapEM'
args['dataset_name'] = dataset_name
args['cell_size'] = cell_size
args['hidden_caps'] = hidden_caps
args['num_layers'] = num_layers
args['epochs'] = epochs
args['batch_size'] = batch_size
args['l_rate'] = l_rate
args['train'] = 0 if restore == 1 else 1
args['results'] = 1
args['plot'] = 1
args['restore'] = restore
args['early_stopping'] = 1
coords = list()
d = cell_size
for i, l in enumerate(range(num_layers)):
capLayer = list()
for j, c in enumerate(range(hidden_caps)):
if j == 0:
capLayer.append([d, (d + j * d + i * d) // 2])
else:
capLayer.append([(d + j * d + i * d) // 2,
(d + i * d) // 2])
coords.append(capLayer)
args['coords'] = coords
dirs = ['checkpoint_dir', 'summary_dir', 'result_dir', 'log_dir']
for dr in dirs:
args[dr] = dr
self.config = utils.Config(args)
def build(self):
''' ------------------------------------------------------------------------------
SET NETWORK PARAMS
------------------------------------------------------------------------------ '''
network_params_dict = dict()
network_params_dict['input_height'] = self.data_train.height
network_params_dict['input_width'] = self.data_train.width
network_params_dict['input_nchannels'] = self.data_train.num_channels
network_params_dict['train_size'] = self.data_train._ndata
network_params_dict['num_classes'] = self.data_train._labels.shape[1]
network_params_dict['cell_size'] = self.config.cell_size
network_params_dict['hidden_caps'] = self.config.hidden_caps
network_params_dict['num_layers'] = self.config.num_layers
self.network_params = utils.Config(network_params_dict)
self._build()
def build(self):
''' ------------------------------------------------------------------------------
SET NETWORK PARAMS
------------------------------------------------------------------------------ '''
network_params_dict = dict()
network_params_dict['input_height'] = self.data_train.height
network_params_dict['input_width'] = self.data_train.width
network_params_dict['input_nchannels'] = self.data_train.num_channels
network_params_dict['train_size'] = self.data_train._ndata
network_params_dict['hidden_dim'] = self.config.hidden_dim
network_params_dict['latent_dim'] = self.config.latent_dim
network_params_dict['l2'] = self.config.l2
network_params_dict['num_layers'] = self.config.num_layers
self.network_params = utils.Config(network_params_dict)
self._build()