# Set optimizer.
optimizer_conf = dict(param_interface['training']['optimizer'])
optimizer_name = optimizer_conf['name']
del optimizer_conf['name']
# Select for optimization only those parameters that require update!
optimizer = getattr(torch.optim,
optimizer_name)(filter(lambda p: p.requires_grad,
model.parameters()),
**optimizer_conf)
# Ok, finished loading the configuration.
# Save the resulting configuration into a yaml settings file, under log_dir
with open(log_dir + "training_configuration.yaml",
'w') as yaml_backup_file:
yaml.dump(param_interface.to_dict(),
yaml_backup_file,
default_flow_style=False)
# Log the training configuration.
conf_str = '\n' + '=' * 80 + '\n'
conf_str += 'Final registry configuration for training {} on {}:\n'.format(
model_name, task_name)
conf_str += '=' * 80 + '\n'
conf_str += yaml.safe_dump(param_interface.to_dict(),
default_flow_style=False)
conf_str += '=' * 80 + '\n'
logger.info(conf_str)
# Ask for confirmation - optional.
if FLAGS.confirm:
problem = ProblemFactory.build_problem(
param_interface['testing']['problem'])
# Create statistics collector.
stat_col = StatisticsCollector()
# Add model/problem dependent statistics.
problem.add_statistics(stat_col)
model.add_statistics(stat_col)
# Create test output csv file.
test_file = stat_col.initialize_csv_file(log_dir, 'testing.csv')
# Ok, finished loading the configuration.
# Save the resulting configuration into a yaml settings file, under log_dir
with open(log_dir + "testing_configuration.yaml", 'w') as yaml_backup_file:
yaml.dump(param_interface.to_dict(),
yaml_backup_file, default_flow_style=False)
# Run test
with torch.no_grad():
for episode, (data_tuple, aux_tuple) in enumerate(
problem.return_generator()):
if episode == param_interface["testing"]["problem"][
"max_test_episodes"]:
break
logits, loss = forward_step(
model, problem, episode, stat_col, data_tuple, aux_tuple)
# Log to logger.