为模仿框架启用 Tensorboard 和奖励记录

Question

CHAI是加州大学伯克利分校的模仿和奖励学习算法（逆RL）框架>（https://github.com/HumanCompatibleAI/imitation）

我无法为

imitation

库启用日志记录和 Tensorboard。这是

logger

实用程序 - 感谢任何帮助。

https://imitation.readthedocs.io/en/latest/_api/imitation.util.logger.html

Answer 1

我弄清楚了这个机制并将其发布在这里供其他可能感兴趣的人：

from imitation.algorithms import bc
from imitation.util import logger as imit_logger

# Set new logger
log_path = '/irl_log/'
logger_irl = imit_logger.configure(log_path, ["stdout", "csv", "tensorboard"])

bc_trainer = bc.BC(
    observation_space=mt_venv.observation_space,
    action_space=mt_venv.action_space,
    demonstrations=transitions,
    rng=rng,
    device='cpu',
    custom_logger = logger_irl 
)