fgail3456 / fgail Goto Github PK

Dear author,
I notice that you use an nn to estimate f-conjugate. In my opinion, for RKL, KL, JS divergence, the form of f-conjugate is given. And why you use a nn instead?

looking for your reply.
Best, Xiaoyuan

Hi,

I am attempting to run the code as stated in the README, but I am running into some Module Not Found Errors. There seems to be some missing Python files, could you upload these files as well, please? Thanks.

$ python3 main.py --env-name HalfCheetah-v2 --algo ppo --use-gae --lr 3e-4 --clip-param 0.1 --value-loss-coef 0.5 --num-processes 8 --num-steps 1000 --num-mini-batch 4 --log-interval 1 --use-linear-lr-decay --entropy-coef 0.01
Traceback (most recent call last):
  File "main.py", line 20, in <module>
    from a2c_ppo_acktr.model import Policy
ModuleNotFoundError: No module named 'a2c_ppo_acktr.model'

From main.py:

from a2c_ppo_acktr.model import Policy
from a2c_ppo_acktr.storage import RolloutStorage
from evaluation import evaluate

are you available for freelancing?
what is your email?

Hi, I notice you use 'type' in 'args' to set different kinds of divergence. But in the f-gail script. I notice you don't explicitly set a different kind of divergence. If I wanna use a specific kind of divergence. How to set?

Best,
Yuechao.