TextWorld
TextWorld copied to clipboard
microsoft
Steps with multiple commands in them will crash the gym
Inform's parser allows multiple commands to be expressed on a single line, e.g. "GO NORTH. GO EAST." or "DROP PEPPER THEN GET KNIFE". Calling env.step with such an argument will cause an exception in Inform7Data._gather_infos.
Interesting. Do you have a minimal working example I can use to debug (i.e., the generated files .json and .z8)?
Looking back at TextWorld's code, it should be able to parse multiple commands (at least part of the codebase is able to): https://github.com/microsoft/TextWorld/blob/main/textworld/envs/wrappers/tw_inform7.py#L300-L313
The attached game was generated via tw-make tw-cooking --recipe 3 --take 2 --go 12 --open --cook --cut --drop --seed 1: ecf1762719a475c1fe7a8fce3c09bddc519a3f598d9afa7ece93f01ca0f096db-1.zip. TextWorld was installed by executing !{sys.executable} -m pip install textworld from a default Colab environment.
import textworld, textworld.gym
def crash_demo(path):
request_infos = textworld.core.EnvInfos(lost=True, won=True)
env_id = textworld.gym.register_game(
path, request_infos=request_infos,
)
game_env = textworld.gym.make(env_id)
game_env.reset()
game_env.step("OPEN SLIDING PATIO DOOR THEN GO NORTH")
Calling crash_demo on this game yields:
---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
[<ipython-input-55-db19323702ae>](https://localhost:8080/#) in <cell line: 0>()
----> 1 crash_demo(make_game(1, TW_MAKE_ARGS))
12 frames
[<ipython-input-50-505e40e69955>](https://localhost:8080/#) in crash_demo(path)
8 game_env.reset()
9
---> 10 game_env.step("OPEN SLIDING PATIO DOOR THEN GO NORTH")
[/usr/local/lib/python3.11/dist-packages/textworld/gym/envs/textworld.py](https://localhost:8080/#) in step(self, command)
63 * infos: additional information as requested.
64 """
---> 65 obs, scores, dones, infos = super().step([command])
66 return obs[0], scores[0], dones[0], {k: v[0] for k, v in infos.items()}
[/usr/local/lib/python3.11/dist-packages/textworld/gym/envs/textworld_batch.py](https://localhost:8080/#) in step(self, commands)
148
149 self.last_commands = commands
--> 150 self.obs, scores, dones, infos = self.batch_env.step(self.last_commands)
151 return self.obs, scores, dones, infos
152
[/usr/local/lib/python3.11/dist-packages/textworld/envs/batch/batch_env.py](https://localhost:8080/#) in step(self, actions)
269 results.append((obs, reward, done, infos))
270 else:
--> 271 results.append(env.step(action))
272
273 self.last = results
[/usr/local/lib/python3.11/dist-packages/textworld/envs/wrappers/filter.py](https://localhost:8080/#) in step(self, command)
44
45 def step(self, command: str) -> Tuple[str, Mapping[str, Any]]:
---> 46 game_state, score, done = super().step(command)
47 ob = game_state.feedback
48 infos = self._get_requested_infos(game_state)
[/usr/local/lib/python3.11/dist-packages/textworld/core.py](https://localhost:8080/#) in step(self, command)
347
348 def step(self, command: str) -> Tuple[GameState, float, bool]:
--> 349 return self._wrapped_env.step(command)
350
351 def reset(self) -> GameState:
[/usr/local/lib/python3.11/dist-packages/textworld/envs/wrappers/limit.py](https://localhost:8080/#) in step(self, command)
25
26 def step(self, command: str) -> Tuple[str, Mapping[str, Any]]:
---> 27 game_state, score, done = super().step(command)
28 self.nb_steps += 1
29 done |= self.nb_steps >= self.max_episode_steps
[/usr/local/lib/python3.11/dist-packages/textworld/core.py](https://localhost:8080/#) in step(self, command)
347
348 def step(self, command: str) -> Tuple[GameState, float, bool]:
--> 349 return self._wrapped_env.step(command)
350
351 def reset(self) -> GameState:
[/usr/local/lib/python3.11/dist-packages/textworld/core.py](https://localhost:8080/#) in step(self, command)
347
348 def step(self, command: str) -> Tuple[GameState, float, bool]:
--> 349 return self._wrapped_env.step(command)
350
351 def reset(self) -> GameState:
[/usr/local/lib/python3.11/dist-packages/textworld/core.py](https://localhost:8080/#) in step(self, command)
347
348 def step(self, command: str) -> Tuple[GameState, float, bool]:
--> 349 return self._wrapped_env.step(command)
350
351 def reset(self) -> GameState:
[/usr/local/lib/python3.11/dist-packages/textworld/envs/wrappers/tw_inform7.py](https://localhost:8080/#) in step(self, command)
293
294 def step(self, command: str):
--> 295 self.state, score, done = self._wrapped_env.step(command)
296 if not self.tracking:
297 return self.state, score, done # State tracking not needed.
[/usr/local/lib/python3.11/dist-packages/textworld/envs/wrappers/tw_inform7.py](https://localhost:8080/#) in step(self, command)
146 extra_infos, self.state["feedback"] = _detect_extra_infos(self.state["feedback"], self._tracked_infos)
147 self.state.update(extra_infos)
--> 148 self._gather_infos()
149 self.state["done"] = self.state["won"] or self.state["lost"]
150 return self.state, self.state["score"], self.state["done"]
[/usr/local/lib/python3.11/dist-packages/textworld/envs/wrappers/tw_inform7.py](https://localhost:8080/#) in _gather_infos(self)
136 for info in ["score", "moves"]:
137 if self.state[info] is not None and type(self.state[info]) is not int:
--> 138 self.state[info] = int(self.state[info].strip())
139
140 self.state["won"] = '*** The End ***' in self.state["feedback"]
ValueError: invalid literal for int() with base 10: '1\n</moves>\n-= Corridor =-\nYou find yourself in a corridor. A typical kind of place. You try to gain information on your surroundings by using a technique you call "looking."\n\n\n\nThere is an ope