train_mpe_qmix.sh

#!/bin/sh
env="MPE"
scenario="simple_spread" # aka cooperative navigation
num_landmarks=3
num_agents=3
algo="qmix"
exp="${exp:-debug}" # default to experiment name "debug"
seed_max=1
# WARNING: make device num is correct, e.g. check avail with nvidid-smi
CUDA_VISIBLE_DEVICES=0
PYTHON_BIN=../../env/bin/python3
echo "python bin path is ${PYTHON_BIN}"
echo "env is ${env}, scenario is ${scenario}, algo is ${algo}, exp is ${exp}, max seed is ${seed_max}"
if [ "${use_wandb}" = "true" ]; then
    wandb_flag="--use_wandb"
else
    wandb_flag=""
fi
for seed in $(seq ${seed_max}); do
    echo "seed is ${seed}:"
    CUDA_VISIBLE_DEVICES=${CUDA_VISIBLE_DEVICES} ${PYTHON_BIN} 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} \
        --episode_length 25 \
        --batch_size 32 \
        --tau 0.005 \
        --lr 7e-4 \
        --hard_update_interval_episode 100 \
        --num_env_steps 10000000 \
        --use_reward_normalization \
        --user_name elles \
        --fov -1 \
        ${wandb_flag}
    echo "training is done!"
done