__init__() got multiple values for argument 'device'

[colourfulspring]

I run ./onpolicy/scripts/train_mpe_scripts/train_mpe_spread.sh after change 'algo' to mappo and user_name to my wandb user name in train_mpe_spread.sh. My train_mpe_spread.sh is as follows:

#!/bin/sh
env="MPE"
scenario="simple_spread" 
num_landmarks=3
num_agents=3
algo="mappo" #"rmappo" "ippo"
exp="check"
seed_max=1

echo "env is ${env}, scenario is ${scenario}, algo is ${algo}, exp is ${exp}, max seed is ${seed_max}"
for seed in `seq ${seed_max}`;
do
    echo "seed is ${seed}:"
    CUDA_VISIBLE_DEVICES=0 python ../train/train_mpe.py --env_name ${env} --algorithm_name ${algo} --experiment_name ${exp} \
    --scenario_name ${scenario} --num_agents ${num_agents} --num_landmarks ${num_landmarks} --seed ${seed} \
    --n_training_threads 1 --n_rollout_threads 128 --num_mini_batch 1 --episode_length 25 --num_env_steps 20000000 \
    --ppo_epoch 10 --use_ReLU --gain 0.01 --lr 7e-4 --critic_lr 7e-4 --wandb_name "xxx" --user_name "my_wandb_name"
done

Then I got an error

Traceback (most recent call last):
  File "/home/drl/Projects/on-policy/onpolicy/scripts/train_mpe_scripts/../train/train_mpe.py", line 174, in <module>
    main(sys.argv[1:])
  File "/home/drl/Projects/on-policy/onpolicy/scripts/train_mpe_scripts/../train/train_mpe.py", line 158, in main
    runner = Runner(config)
  File "/home/drl/Projects/on-policy/onpolicy/runner/shared/mpe_runner.py", line 14, in __init__
    super(MPERunner, self).__init__(config)
  File "/home/drl/Projects/on-policy/onpolicy/runner/shared/base_runner.py", line 80, in __init__
    self.policy = Policy(self.all_args,
TypeError: __init__() got multiple values for argument 'device'

I found on-policy/onpolicy/runner/shared/base_runner.py, line 66~71 choose R_MAPPOPolicy as Policy:

        if self.algorithm_name == "mat" or self.algorithm_name == "mat_dec":
            from onpolicy.algorithms.mat.mat_trainer import MATTrainer as TrainAlgo
            from onpolicy.algorithms.mat.algorithm.transformer_policy import TransformerPolicy as Policy
        else:
            from onpolicy.algorithms.r_mappo.r_mappo import R_MAPPO as TrainAlgo
            from onpolicy.algorithms.r_mappo.algorithm.rMAPPOPolicy import R_MAPPOPolicy as Policy

The constructor of R_MAPPOPolicy is

    def __init__(self, args, obs_space, cent_obs_space, act_space, device=torch.device("cpu")):

but on-policy/onpolicy/runner/shared/base_runner.py, line 80~85 provide incompatible arguments to the constructor of R_MAPPOPolicy. The second last one is redundant.

        self.policy = Policy(self.all_args,
                            self.envs.observation_space[0],
                            share_observation_space,
                            self.envs.action_space[0],
                            self.num_agents, # default 2
                            device = self.device)

What should I do? Thanks!

[itstyren]

Same issue It appears to be related to a mismatch in the passed parameters. One potential, albeit unsafe, solution would be to simply removeself.num_agents.

[colourfulspring]

OK, thanks!

I found another piece of code similar with this problem. Code snippet onpolicy/runner/shared/base_runner.py, line 90~93 are as follows

        if self.algorithm_name == "mat" or self.algorithm_name == "mat_dec":
            self.trainer = TrainAlgo(self.all_args, self.policy, self.num_agents, device = self.device)
        else:
            self.trainer = TrainAlgo(self.all_args, self.policy, device = self.device)

The code define TrainAlgo together with on-policy/onpolicy/runner/shared/base_runner.py, line 66~71 are

        if self.algorithm_name == "mat" or self.algorithm_name == "mat_dec":
            from onpolicy.algorithms.mat.mat_trainer import MATTrainer as TrainAlgo
            from onpolicy.algorithms.mat.algorithm.transformer_policy import TransformerPolicy as Policy
        else:
            from onpolicy.algorithms.r_mappo.r_mappo import R_MAPPO as TrainAlgo
            from onpolicy.algorithms.r_mappo.algorithm.rMAPPOPolicy import R_MAPPOPolicy as Policy

The TrainAlgo class has a very similar definition compared with the Policy class. Here the author just use an if statement to solve this problem. I think the aforementioned problem can be solved by add an if statement at the calling of Policy's constructor, too. Maybe one possible solution is changing on-policy/onpolicy/runner/shared/base_runner.py, line 80~85 to

        if self.algorithm_name == "mat" or self.algorithm_name == "mat_dec":
            self.policy = Policy(self.all_args, self.envs.observation_space[0], share_observation_space, self.envs.action_space[0], self.num_agents, device = self.device)
        else:
            self.policy = Policy(self.all_args, self.envs.observation_space[0], share_observation_space, self.envs.action_space[0], device = self.device)