def __init__(self) -> None:
self._logger = create_logger()
self._create_api()
self.metadata_manager = MetadataManager.get_instance()
self.run_loop = True
self._logger.info("Manager set up and waiting for data.")
self.run()
def __init__(self,
working_directories: Dict[Type[theory_base.TheoryBase], str],
keep_full_history: bool = False):
"""
:param working_directories: Mapping from theory classes to working directory.
Current directory will be changed accordingly.
:param keep_full_history: whether to keep the full history for each (env, theory) pair or not
"""
self._logger = logger_config.create_logger('rs')
self._best_results = {
} # Dict[Tuple[Type[env_base.EnvironmentBase], Type[theory_base.TheoryBase]], theory_base.TheoryBase]
self._logger.info((
'Creating RoboScientist object with the following configurations:\n'
'working directories: {}\nkeep_full_history: {}').format(
working_directories, keep_full_history))
self._working_directories = working_directories
self._keep_full_history = keep_full_history
self._history = {} if self._keep_full_history else None
parser.add_argument("-p", type=int, default=1)
parser.add_argument("--ntrain", type=int, default=1000)
parser.add_argument("--seed", type=int, default=0)
parser.add_argument(
"--root_data",
type=str,
default="/private/home/asablayrolles/code/projects/whitebox_blackbox/")
parser.add_argument("--normed", type=boolify, default=True)
return parser
if __name__ == "__main__":
parser = get_parser()
args = parser.parse_args()
logger = create_logger()
state = np.random.RandomState(args.seed)
# Loading data
X = np.load(join(args.root_data, "cifar10_cnn.npy"))
label = np.load(join(args.root_data,
"cifar10_label.npy")).astype(int).ravel()
if args.normed:
X /= np.linalg.norm(X, axis=1, keepdims=True)
# Shuffling data
indices = np.nonzero(label >= 8)[0]
indices = indices[state.permutation(indices.shape[0])]
X = X[indices]
y = (label[indices] == 9).astype(int)
# random seed
import torch
import numpy as np
np.random.seed(cfg["seed"])
torch.manual_seed(cfg["seed"])
assert cfg["cuda"], "only support gpu"
if cfg["cuda"]:
torch.cuda.manual_seed(cfg["seed"])
import time
timestamp = time.strftime('%Y-%m-%d-%H-%M')
# create logger
import logging
from lib.logger import create_logger
_ = create_logger(cfg["output_path"], cfg["comment"], timestamp)
# backup config
import shutil
import os
shutil.copyfile(
args.cfg,
os.path.join(cfg["output_path"], "{}_{}".format(timestamp,
cfg["config_fn"])))
# set env
os.environ["CUDA_VISIBLE_DEVICES"] = ",".join(
map(lambda x: str(x), cfg["gpu_ids"]))
from optimizer.adam import Adam
from eval import cal_avg_auc
from lib.user_pigeonhole import save_tag
def parse_args():
parser = argparse.ArgumentParser(description='Train setting')
parser.add_argument('--cfg',
help='configure file name',
required=True,
type=str) # 命令行里包含config文件位置
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parse_args()
update_config(cfg, args) # 从config文件更新配置
logger, _ = create_logger(cfg)
logger.info(cfg)
if cfg.TYPE == 'tag':
for tag in cfg.TAG:
save_tag(cfg, str(tag))
elif cfg.TYPE == 'USER':
for user in cfg.USER:
person_blog(cfg, str(user))
else:
logger.info('=> failed in cfg TYPE. Please enter correct configure.')