continuous action space about on-policy HOT 11 CLOSED

I have the same problem, has anyone solved it?

from on-policy.

I have the same problem, has anyone solved it?

it should be modified.

from on-policy.

but what to change? the codes below only need x

from on-policy.

but what to change? the codes below only need x
maybe:

def forward(self, x, a=None):

from on-policy.

but what to change? the codes below only need x
maybe:
python '''
def forward(self, x, a=None):
'''

sorry, it doesn't work

Traceback (most recent call last):
File "/home/qss/syo/onpolicy/scripts/train/train_mpe.py", line 169, in
main(sys.argv[1:])
File "/home/qss/syo/onpolicy/scripts/train/train_mpe.py", line 154, in main
runner.run()
File "/home/qss/syo/onpolicy/runner/shared/mpe_runner.py", line 28, in run
values, actions, action_log_probs, rnn_states, rnn_states_critic, actions_env = self.collect(step)
File "/home/qss/.conda/envs/syo/lib/python3.6/site-packages/torch/autograd/grad_mode.py", line 15, in decorate_context
return func(*args, **kwargs)
File "/home/qss/syo/onpolicy/runner/shared/mpe_runner.py", line 121, in collect
raise NotImplementedError
NotImplementedError

from on-policy.

The code above also needs to be changed

# raise NotImplementedError
actions_env = actions

from on-policy.

@sinizu would you mind give me your emal please. there are still some problems i want to talk with you.

from on-policy.

You can develop it just like discrete action implement. But i can not get satisfied result when i use the modified code.

from on-policy.

@sinizu would you mind give me your emal please. there are still some problems i want to talk with you.

You can leave yours. I'll text you when I'm free

from on-policy.

You can leave yours. I'll text you when I'm free

thank a lot! I've successfully changed it to continuous action space. however the result is slightly inferior compared with discrete version.

from on-policy.

@Shenyiou
would you mind share your code or tell what you have changed? That would help a lot

from on-policy.