Hi, I have this error in running!

[BCWang93]

The error is like this:

'Process Process-3:
Traceback (most recent call last):
File "/home/a/anaconda3/lib/python3.6/multiprocessing/process.py", line 258, in _bootstrap
self.run()
File "/home/a/anaconda3/lib/python3.6/multiprocessing/process.py", line 93, in run
self._target(*self._args, **self._kwargs)
File "/home/a/COMA/pymarl/src/runners/parallel_runner.py", line 221, in env_worker
reward, terminated, env_info = env.step(actions)
File "/home/a/anaconda3/lib/python3.6/site-packages/smac/env/starcraft2/starcraft2.py", line 373, in step
agent_action = self.get_agent_action(a_id, action)
File "/home/a/anaconda3/lib/python3.6/site-packages/smac/env/starcraft2/starcraft2.py", line 439, in get_agent_action
"Agent {} cannot perform action {}".format(a_id, action)
AssertionError: Agent 0 cannot perform action 0
'
Can you help me to solve this?Thanks!

[PaLeroy]

Hey,
Just got the same error on several runs using COMA.
Were you using COMA too?
However, it never occurred using QMIX algorithms.

[BCWang93]

Hey,
Just got the same error on several runs using COMA.
Were you using COMA too?
However, it never occurred using QMIX algorithms.

yes,I also use the coma.Did you know how to solve this problem?

[BCWang93]

Hey,
Just got the same error on several runs using COMA.
Were you using COMA too?
However, it never occurred using QMIX algorithms.

yes, I also using COMA,Do you know how to solve this problem?

[PaLeroy]

I have not investigated yet.
I got this bug "randomly" after 242030, 10046672 and 4677522 transitions.
It seems to be always caused by the action 0 (the no-op action when a unit is dead).
But the exception occurs simply if the selected action is not an available one, not because the unit is not dead...

[tabzraz]

Sorry for the late reply. It's unlikely I'll be able to take a look at this issue in the next couple of weeks.

Are you using pytorch version >1?

If the problem only occurs with COMA, then the code for the parallel_runner is unlikely to be the problem.

Any unavailable actions should not be chosen by the action selector, but it would be worth checking that is indeed true.

[BCWang93]

Sorry for the late reply. It's unlikely I'll be able to take a look at this issue in the next couple of weeks.

Are you using pytorch version >1?

If the problem only occurs with COMA, then the code for the parallel_runner is unlikely to be the problem.

Any unavailable actions should not be chosen by the action selector, but it would be worth checking that is indeed true.

Hi,thanks for your reply!I am use the pytorch version is >1.I am not test other algorithm recently!But somebody test the qmix algorithm is running normal!

[BCWang93]

Sorry for the late reply. It's unlikely I'll be able to take a look at this issue in the next couple of weeks.
Are you using pytorch version >1?
If the problem only occurs with COMA, then the code for the parallel_runner is unlikely to be the problem.
Any unavailable actions should not be chosen by the action selector, but it would be worth checking that is indeed true.

so,can you solve this problem in recent weeks?Thank you!

[BCWang93]

so,can you solve this problem in recent weeks?Thank you! @samvelyan

[jyericlin]

Encountered the same issue when running QMIX as well. It occurs when running the following command:
python3 src/main.py --config=qmix_smac --env-config=sc2 with env_args.map_name=2s3z

Tested it in two machines, but both have the same issue. I am using PyTorch 0.4.1. I would appreciate if you could provide pointers on how to fix this issue.

[tabzraz]

I'll take another look at this issue soon.
When I've been running experiments using the parallel_runner I haven't come across this.

[tabzraz]

I haven't been able to reproduce this to debug it.

[GoingMyWay]

Same issue with QMIX. Rarely happens but happened today.

AssertionError: Agent 3 cannot perform action 0