document play.sh options

scripts/play.sh

@@ -14,22 +14,22 @@ while [ "$1" != "" ]; do
             ;;
         -h | --header )
             header=1
-            ;;
+            ;; # adds banner with env metadata like action
         -r | --reconstruction )
             reconstruction=1
-            ;;
+            ;; # 3 panes [original_obs, resized_obs, reconstructed], doesn't do anything if any of -w -a or -e are set. shows quality of encoder decoder
         -s | --save-mode )
             save_mode=1
-            ;;
+            ;; # lets you save the episode to mp4
         -a | --agent-world-model )
             mode="agent_in_world_model"
-            ;;
+            ;; # the agent plays in the world model env, shows the quality of the dynamics model
         -e | --episode )
             mode="episode_replay"
-            ;;
+            ;; # replay train, test, or imagined episodes. shows quality of dynamics model
         -w | --world-model )
             mode="play_in_world_model"
-            ;;
+            ;; # human plays in world model
         * )
             echo Invalid usage : $1
             exit 1
@@ -37,4 +37,4 @@ while [ "$1" != "" ]; do
     shift
 done
 
-python src/play.py hydra.run.dir=. hydra.output_subdir=null +mode="${mode}" +fps="${fps}" +header="${header}" +reconstruction="${reconstruction}" +save_mode="${save_mode}"
+python -m pdb src/play.py hydra.run.dir=. hydra.output_subdir=null +mode="${mode}" +fps="${fps}" +header="${header}" +reconstruction="${reconstruction}" +save_mode="${save_mode}"

src/envs/wrappers.py

@@ -39,6 +39,7 @@ def make_crafter(id, size=64, max_episode_steps=None, done_on_life_loss=False):
     # https://github.com/danijar/dreamerv2/blob/07d906e9c4322c6fc2cd6ed23e247ccd6b7c8c41/dreamerv2/common/envs.py#L242
     # https://github.com/footoredo/torchbeast/blob/12939569cc46b6a8616e4c25b138d97248cc8581/torchbeast/atari_wrappers.py#L301
     env = gym.make(id)
+    env = ResizeObsWrapper(env, (size, size))
     return env
 
 

src/game/agent_env.py

@@ -4,14 +4,14 @@ from PIL import Image
 import torch
 from torchvision.transforms.functional import InterpolationMode, resize
 
-from agent import Agent
-from envs import SingleProcessEnv, WorldModelEnv
-from game.keymap import get_keymap_and_action_names
+from src.agent import Agent
+from src.envs import SingleProcessEnv, WorldModelEnv
+from src.game.keymap import get_keymap_and_action_names
 
 
 class AgentEnv:
     def __init__(self, agent: Agent, env: SingleProcessEnv, keymap_name: str, do_reconstruction: bool) -> None:
-        assert isinstance(env, SingleProcessEnv) or isinstance(env, WorldModelEnv)
+        assert isinstance(env, SingleProcessEnv) or isinstance(env, WorldModelEnv), f"{env}"
         self.agent = agent
         self.env = env
         _, self.action_names = get_keymap_and_action_names(keymap_name)