Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

How come different performance? #12

Open
tessavdheiden opened this issue Jan 30, 2020 · 5 comments
Open

How come different performance? #12

tessavdheiden opened this issue Jan 30, 2020 · 5 comments

Comments

@tessavdheiden
Copy link

Hi Morvan,

Why do I get totally different performance (see file attached)?

Best,
Tessa
a3c_continious

@ghost
Copy link

ghost commented Mar 11, 2020

Hi,

How about changing the 'UPDATE_GLOBAL_ITER' more than 5?

It was helpful to me and I got following performance with UPDATE_GLOBAL_ITER=10.

image

@pengzhi1998
Copy link

Hi, but when I ran this code, the moving average reward is always below -1000 for the continuous situation, do you know what kind of problem it could be? (the 'UPDATE_GLOBAL_ITER' has already been set to 10) The performance of the discrete situation is very bad as well.

@ghost
Copy link

ghost commented Mar 14, 2020

Hi,

Here is another trial.

Try 'torch.nn.utils.clip_grad_norm_(lnet.parameters(), 20)' in utils.py

image

It helped me to reduce performance differences.

@pengzhi1998
Copy link

Hi,

Here is another trial.

Try 'torch.nn.utils.clip_grad_norm_(lnet.parameters(), 20)' in utils.py

image

It helped me to reduce performance differences.

Thank you! I'll take a try.

@Eify666666
Copy link

Hi,

Here is another trial.

Try 'torch.nn.utils.clip_grad_norm_(lnet.parameters(), 20)' in utils.py

image

It helped me to reduce performance differences.

Hi,I meet a trouble when I train another A3C.
After some time, all the networks always output the same action.
I tried the "torch.nn.utils.clip_grad_norm_(lnet.parameters(), 20)", it doesn't work.
It may be that during the training process, the network tries many times, but does not reap the reward.
Do you have any ideas about this problem?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants