configs.append(parser.parse_args())
return configs
if __name__ == "__main__":
experiments = argparse.ArgumentParser()
experiments.add_argument('--config_path',
default='experiments.json',
type=str,
help='Path of experiments config file')
experiments = experiments.parse_args()
from utils.Pytorch_GPUManager import GPUManager
index = GPUManager().auto_choice()
device = torch.device("cuda:" +
str(index) if torch.cuda.is_available() else "cpu")
exp_configs = parse_experiments(experiments.config_path)
n = 5
for config in exp_configs:
logger.info('-' * 80)
logger.info('Config {} (totally {} configs)'.format(
exp_configs.index(config) + 1, len(exp_configs)))
results = []
max_apc_test_acc, max_apc_test_f1, max_ate_test_f1 = 0, 0, 0
for i in range(n):
config.device = device
config.seed = i + 1
logger.info('No.{} training process of {}'.format(i + 1, n))
apc_test_acc, apc_test_f1, ate_test_f1 = main(config)
mean_test_acc_overall = temp_test_acc_overall / (t + 1)
mean_f1_overall = temp_f1_overall / (t + 1)
logging.info('max_acc_overall:{} f1_overall:{}'.format(round(max_acc_overall, 2), round(max_f1_overall, 2)))
logging.info("mean_acc_overall:{} mean_f1_overall:{}".format(round(mean_test_acc_overall, 2),
round(mean_f1_overall, 2)))
logging.info("#" * 100)
return mean_test_acc_overall, mean_f1_overall
if __name__ == '__main__':
config_parser = argparse.ArgumentParser()
config_parser.add_argument('--config', default='experiments_apc.json',
help='path of the experiments configuration', type=str)
args = config_parser.parse_args()
configs = parse_experiments(args.config)
log_file = 'logs/{}.{}.log'.format(args.config, strftime("%y%m%d.%H%M", localtime()))
logger.addHandler(logging.FileHandler(log_file))
from utils.Pytorch_GPUManager import GPUManager
GM = GPUManager()
gpu = GM.auto_choice()
for config in configs:
config.device = 'cuda:' + str(gpu)
# config.device = 'cpu' # Uncomment this line to use CPU
multi_train(config=config, n=config.repeat)