add script for training qmix or mqmix

offpolicy/scripts/train_custom_qmix.sh

@@ -1,10 +1,10 @@
 #!/bin/sh
-env="custom"
-scenario="shared"
+env="car"
+scenario="shared_7a"
 num_landmarks=3
 num_agents=4
 algo="mqmix"
-exp="luoming"
+exp="el1024_bas32_nr16_use_per"
 seed_max=1
 
 current_dir=$(cd $(dirname $0); pwd)
@@ -15,9 +15,9 @@ for seed in $(seq ${seed_max}); do
     echo "seed is ${seed}:"
     CUDA_VISIBLE_DEVICES=7 python -u ${current_dir}/train/train_custom.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 2048 \
-    --batch_size 32 --lr 7e-4 --hard_update_interval_episode 100 --num_env_steps 100000000 \
-    --n_rollout_threads 16 --buffer_size 10000 --save_interval 50000 --epsilon_anneal_time 50000 \
-    --train_interval 100 --use_reward_normalization --use_wandb --cuda
+    --num_agents ${num_agents} --num_landmarks ${num_landmarks} --seed ${seed} --episode_length 1024 \
+    --batch_size 128 --lr 5e-4 --num_env_steps 300000000 \
+    --n_rollout_threads 16 --buffer_size 10000 --save_interval 50000 --epsilon_anneal_time 500000 \
+    --train_interval 256 --num_eval_episodes 1 --use_wandb --cuda --use_per
     echo "training is done!"
 done

offpolicy/scripts/train_custom_qmix_rnn.sh

@@ -0,0 +1,23 @@
+#!/bin/sh
+env="custom"
+scenario="shared_7a"
+num_landmarks=3
+num_agents=4
+algo="qmix"
+exp="el1024_bas32"
+seed_max=1
+
+current_dir=$(cd $(dirname $0); pwd)
+
+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=7 python -u ${current_dir}/train/train_custom.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 1024 \
+    --batch_size 32 --lr 5e-4 --num_env_steps 300000000 \
+    --n_rollout_threads 1 --buffer_size 5000 --save_interval 50000 --epsilon_anneal_time 500000 \
+    --train_interval_episode 1 --num_eval_episodes 1 --use_wandb
+    echo "training is done!"
+done