def get_loggers(self):
loggers = {}
labels = ['input', 'output', 'attn']
for key in labels:
loggers[key] = LazyRegisterer(
os.path.join(self.logs_folder,
'{}_{}.png'.format(key, self.split)), 'image',
'Samples {} {}'.format(key, self.split))
return loggers
def get_loggers(self, add_orientation, split):
loggers = {}
labels = ['input', 'gt_segmentation', 'output_segmentation']
if add_orientation:
labels.extend(['gt_orientation', 'output_orientation'])
for name in labels:
key = '{}_{}'.format(name, split)
loggers[name] = LazyRegisterer(
os.path.join(self.logs_folder, '{}.png'.format(key)), 'image',
'Samples {} {}'.format(name, split))
return loggers
def _get_plot_loggers(model_opt, train_opt):
samples = {}
_ssets = ['train', 'valid']
for _set in _ssets:
labels = ['input', 'output']
for name in labels:
key = '{}_{}'.format(name, _set)
samples[key] = LazyRegisterer(
os.path.join(logs_folder, '{}.png'.format(key)), 'image',
'Samples {} {}'.format(name, _set))
return samples
def get_loggers(self):
loggers = {}
labels = ['input', 'output', 'total', 'box', 'patch', 'attn']
if self.model_opt['add_d_out']:
labels.append('d_in')
if self.model_opt['add_y_out']:
labels.append('y_in')
for key in labels:
loggers[key] = LazyRegisterer(
os.path.join(self.logs_folder, '{}_{}.png'.format(key, self.split)),
'image', 'Samples {} {}'.format(key, self.split))
return loggers
def get_plot_loggers(model_opt, train_opt):
samples = {}
_ssets = ['train', 'valid']
for _set in _ssets:
labels = ['input', 'output']
if model_opt['ctrl_rnn_inp_struct'] == 'attn':
labels.append('attn')
for name in labels:
key = '{}_{}'.format(name, _set)
samples[key] = LazyRegisterer(
os.path.join(logs_folder, '{}.png'.format(key)), 'image',
'Samples {} {}'.format(name, _set))
return samples
def get_plot_loggers(model_opt, train_opt):
samples = {}
_ssets = ['train']
if train_opt['has_valid']:
_ssets.append('valid')
for _set in _ssets:
labels = ['input', 'output', 'total']
if model_opt['type'] == 'attention':
num_ctrl_cnn = len(model_opt['ctrl_cnn_filter_size'])
num_attn_cnn = len(model_opt['attn_cnn_filter_size'])
num_attn_dcnn = len(model_opt['attn_dcnn_filter_size'])
labels.extend(['box', 'patch'])
if model_opt['ctrl_rnn_inp_struct'] == 'attn':
labels.append('attn')
for name in labels:
key = '{}_{}'.format(name, _set)
samples[key] = LazyRegisterer(
os.path.join(logs_folder, '{}.png'.format(key)), 'image',
'Samples {} {}'.format(name, _set))
return samples
'train batch mean', 'valid batch mean',
'train batch variance', 'valid batch variance', 'ema mean',
'ema variance'
],
name='D-CNN {} batch norm stats'.format(ii),
buffer_size=1)
dcnn_bn_loggers.append(_dcnn_bn_logger)
log_manager.register(log.filename, 'plain', 'Raw logs')
model_opt_fname = os.path.join(logs_folder, 'model_opt.yaml')
saver.save_opt(model_opt_fname, model_opt)
log_manager.register(model_opt_fname, 'plain', 'Model hyperparameters')
valid_sample_img = LazyRegisterer(
os.path.join(logs_folder, 'valid_sample_img.png'), 'image',
'Validation samples')
train_sample_img = LazyRegisterer(
os.path.join(logs_folder, 'train_sample_img.png'), 'image',
'Training samples')
log.info(
('Visualization can be viewed at: '
'http://{}/deep-dashboard?id={}').format(args.localhost,
model_id))
num_ex_train = dataset['train']['input'].shape[0]
num_ex_valid = dataset['valid']['input'].shape[0]
get_batch_train = _get_batch_fn(dataset['train'])
get_batch_valid = _get_batch_fn(dataset['valid'])
batch_size = args.batch_size
log.info('Number of validation examples: {}'.format(num_ex_valid))