Merge pull request #156 from facebookresearch/eric/competition

Eric/competition

Author: cdmatters

nle/env/__init__.py

@@ -16,5 +16,8 @@
 registration.register(id="NetHackGold-v0", entry_point="nle.env.tasks:NetHackGold")
 registration.register(id="NetHackEat-v0", entry_point="nle.env.tasks:NetHackEat")
 registration.register(id="NetHackScout-v0", entry_point="nle.env.tasks:NetHackScout")
+registration.register(
+    id="NetHackChallenge-v0", entry_point="nle.env.tasks:NetHackChallenge"
+)
 
 __all__ = ["NLE", "DUNGEON_SHAPE"]

nle/env/base.py

@@ -94,7 +94,7 @@
     ),
     (
         "inv_strs",
-        gym.spaces.Box(low=0, high=127, **nethack.OBSERVATION_DESC["inv_strs"]),
+        gym.spaces.Box(low=0, high=255, **nethack.OBSERVATION_DESC["inv_strs"]),
     ),
     (
         "inv_letters",
@@ -116,13 +116,13 @@
     ),
     (
         "tty_chars",
-        gym.spaces.Box(low=0, high=127, **nethack.OBSERVATION_DESC["tty_chars"]),
+        gym.spaces.Box(low=0, high=255, **nethack.OBSERVATION_DESC["tty_chars"]),
     ),
     (
         "tty_colors",
         gym.spaces.Box(
-            low=-15,
-            high=15,
+            low=0,
+            high=31,
             **nethack.OBSERVATION_DESC["tty_colors"],
         ),
     ),
@@ -210,6 +210,7 @@ def __init__(
         options=None,
         wizard=False,
         allow_all_yn_questions=False,
+        allow_all_modes=False,
         space_dict=None,
     ):
         """Constructs a new NLE environment.
@@ -235,11 +236,15 @@ def __init__(
                 If set to True, no y/n questions in step() are declined.
                 If set to False, only elements of SKIP_EXCEPTIONS are not declined.
                 Defaults to False.
+            allow_all_modes (bool):
+                If set to True, do not decline menus, text input or auto 'MORE'.
+                If set to False, only skip click through 'MORE' on death.
         """
 
         self.character = character
         self._max_episode_steps = max_episode_steps
         self._allow_all_yn_questions = allow_all_yn_questions
+        self._allow_all_modes = allow_all_modes
 
         if actions is None:
             actions = FULL_ACTIONS
@@ -339,6 +344,9 @@ def print_action_meanings(self):
         for a_idx, a in enumerate(self._actions):
             print(a_idx, a)
 
+    def _check_abort(self, observation):
+        return self._steps >= self._max_episode_steps
+
     def step(self, action: int):
         """Steps the environment.
 
@@ -360,15 +368,17 @@ def step(self, action: int):
         last_observation = tuple(a.copy() for a in self.last_observation)
 
         observation, done = self.env.step(self._actions[action])
-        observation, done = self._perform_known_steps(
-            observation, done, exceptions=True
-        )
+        is_game_over = observation[self._program_state_index][0] == 1
+        if is_game_over or not self._allow_all_modes:
+            observation, done = self._perform_known_steps(
+                observation, done, exceptions=True
+            )
 
         self._steps += 1
 
         self.last_observation = observation
 
-        if self._steps >= self._max_episode_steps:
+        if self._check_abort(observation):
             end_status = self.StepStatus.ABORTED
         else:
             end_status = self._is_episode_end(observation)

nle/env/tasks.py

@@ -278,3 +278,76 @@ def _reward_fn(self, last_observation, observation, end_status):
         self.dungeon_explored[key] = explored
         time_penalty = self._get_time_penalty(last_observation, observation)
         return reward + time_penalty
+
+
+class NetHackChallenge(NetHackScore):
+    """Environment for the NetHack Challenge.
+
+    The task is an augmentation of the standard NLE task. This is the NLE Score Task
+    but with some subtle differences:
+        * the action space is fixed to include the full keyboard
+        * menus and "<More>" tokens are not skipped
+        * starting character is randomly assigned
+    """
+
+    def __init__(
+        self,
+        *args,
+        character="@",
+        allow_all_yn_questions=True,
+        allow_all_modes=True,
+        penalty_mode="constant",
+        penalty_step: float = -0.00,
+        penalty_time: float = -0.0,
+        max_episode_steps: int = 1e6,
+        observation_keys=(
+            "glyphs",
+            "chars",
+            "colors",
+            "specials",
+            "blstats",
+            "message",
+            "inv_glyphs",
+            "inv_strs",
+            "inv_letters",
+            "inv_oclasses",
+            "tty_chars",
+            "tty_colors",
+            "tty_cursor",
+        ),
+        no_progress_timeout: int = 10_000,
+        **kwargs,
+    ):
+        actions = nethack.ACTIONS
+        super().__init__(
+            *args,
+            actions=actions,
+            character=character,
+            allow_all_yn_questions=allow_all_yn_questions,
+            allow_all_modes=allow_all_modes,
+            penalty_mode=penalty_mode,
+            penalty_step=penalty_step,
+            penalty_time=penalty_time,
+            max_episode_steps=max_episode_steps,
+            observation_keys=observation_keys,
+            **kwargs,
+        )
+        # If the in-game turn count doesn't change for 10_000 steps, we abort
+        self._turns = None
+        self._no_progress_count = 0
+        self.no_progress_timeout = no_progress_timeout
+
+    def _check_abort(self, observation):
+        """Check if time has stopped and no observations has changed long enough
+        to trigger an abort."""
+
+        turns = observation[self._blstats_index][20]
+        if self._turns == turns:
+            self._no_progress_count += 1
+        else:
+            self._turns = turns
+            self._no_progress_count = 0
+        return (
+            self._steps >= self._max_episode_steps
+            or self._no_progress_count >= self.no_progress_timeout
+        )

nle/nethack/actions.py

@@ -14,9 +14,22 @@ def C(c):
     return 0x1F & c
 
 
-# Missing here:
-#   Some characters for text input (e.g., +).
-# General menu handling isn't part of this either.
+class TextCharacters(enum.IntEnum):
+    PLUS = ord("+")
+    MINUS = ord("-")
+    SPACE = ord(" ")
+    APOS = ord("'")
+    QUOTE = ord('"')
+    NUM_0 = ord("0")
+    NUM_1 = ord("1")
+    NUM_2 = ord("2")
+    NUM_3 = ord("3")
+    NUM_4 = ord("4")
+    NUM_5 = ord("5")
+    NUM_6 = ord("6")
+    NUM_7 = ord("7")
+    NUM_8 = ord("8")
+    NUM_9 = ord("9")
 
 
 class CompassCardinalDirection(enum.IntEnum):
@@ -76,6 +89,12 @@ class MiscAction(enum.IntEnum):
     MORE = ord("\r")  # read the next message
 
 
+class UnsafeActions(enum.IntEnum):
+    # currently these result in an error or undesirable behaviour
+    HELP = ord("?")  # give a help message
+    PREVMSG = C("p")  # view recent game messages
+
+
 class Command(enum.IntEnum):
     EXTCMD = ord("#")  # perform an extended command
     EXTLIST = M("?")  # list all extended commands
@@ -100,7 +119,6 @@ class Command(enum.IntEnum):
     FIGHT = ord("F")  # Prefix: force fight even if you don't see a monster
     FORCE = M("f")  # force a lock
     GLANCE = ord(";")  # show what type of thing a map symbol corresponds to
-    HELP = ord("?")  # give a help message
     HISTORY = ord("V")  # show long version and game history
     INVENTORY = ord("i")  # show your inventory
     INVENTTYPE = ord("I")  # inventory specific item types
@@ -121,7 +139,6 @@ class Command(enum.IntEnum):
     PAY = ord("p")  # pay your shopping bill
     PICKUP = ord(",")  # pick up things at the current location
     PRAY = M("p")  # pray to the gods for help
-    PREVMSG = C("p")  # view recent game messages
     PUTON = ord("P")  # put on an accessory (ring, amulet, etc)
     QUAFF = ord("q")  # quaff (drink) something
     QUIT = M("q")  # exit without saving current game
@@ -132,6 +149,7 @@ class Command(enum.IntEnum):
     RIDE = M("R")  # mount or dismount a saddled steed
     RUB = M("r")  # rub a lamp or a stone
     RUSH = ord("g")  # Prefix: rush until something interesting is seen
+    RUSH2 = ord("G")  # Prefix: rush until something interesting is seen
     SAVE = ord("S")  # save the game and exit
     SEARCH = ord("s")  # search for traps and secret doors
     SEEALL = ord("*")  # show all equipment in use
@@ -163,6 +181,7 @@ class Command(enum.IntEnum):
     + list(MiscDirection)
     + list(MiscAction)
     + list(Command)
+    + list(TextCharacters)
 )
 
 NON_RL_ACTIONS = (
@@ -172,13 +191,11 @@ class Command(enum.IntEnum):
     Command.EXTCMD,  # Potentially useful for some wizard actions.
     Command.EXTLIST,
     Command.GLANCE,
-    Command.HELP,
     Command.HISTORY,
     Command.KNOWN,  # Could potentially be useful.
     Command.KNOWNCLASS,  # Could potentially be useful.
     Command.OPTIONS,
     Command.OVERVIEW,  # Could potentially be useful.
-    Command.PREVMSG,  # Could potentially be useful.
     Command.TELEPORT,
     Command.QUIT,
     Command.REDRAW,
@@ -191,13 +208,15 @@ class Command(enum.IntEnum):
 )
 
 _USEFUL_ACTIONS = list(ACTIONS)
-for action in NON_RL_ACTIONS:
+for action in NON_RL_ACTIONS + tuple(TextCharacters):
     _USEFUL_ACTIONS.remove(action)
+_USEFUL_ACTIONS.append(TextCharacters.SPACE)
 USEFUL_ACTIONS = tuple(_USEFUL_ACTIONS)
 del _USEFUL_ACTIONS
 
 _ACTIONS_DICT = {}
 for enum_class in [
+    TextCharacters,
     CompassDirection,
     CompassDirectionLonger,
     MiscDirection,

nle/scripts/play.py

@@ -45,11 +45,12 @@ def get_action(env, action_mode, is_raw_env):
             action = env.action_space.sample()
         else:
             action = random.choice(_ACTIONS)
+            print(action)
     elif action_mode == "human":
         while True:
             with no_echo():
                 ch = ord(os.read(0, 1))
-            if ch in [nethack.C("c"), ord(b"q")]:
+            if ch in [nethack.C("c")]:
                 print("Received exit code {}. Aborting.".format(ch))
                 return None
             try:
@@ -67,7 +68,18 @@ def get_action(env, action_mode, is_raw_env):
     return action
 
 
-def play(env, mode, ngames, max_steps, seeds, savedir, no_render, render_mode, debug):
+def play(
+    env,
+    mode,
+    ngames,
+    max_steps,
+    seeds,
+    savedir,
+    no_render,
+    render_mode,
+    print_frames_separately,
+    **kwargs,
+):
     env_name = env
     is_raw_env = env_name == "raw"
 
@@ -100,10 +112,15 @@ def play(env, mode, ngames, max_steps, seeds, savedir, no_render, render_mode, d
     while True:
         if not no_render:
             if not is_raw_env:
-                print("Previous reward:", reward)
-                if action is not None:
-                    print("Previous action: %s" % repr(env._actions[action]))
+                print("--------")
+                print(f"Previous reward: {str(reward):64s}")
+                act_str = repr(env._actions[action]) if action is not None else ""
+                print(f"Previous action: {str(act_str):64s}")
+                print("--------")
                 env.render(render_mode)
+                print("--------")
+                if not print_frames_separately:
+                    print("\033[31A")  # Go up 31 lines.
             else:
                 print("Previous action:", action)
                 _, chars, _, _, blstats, message, *_ = obs
@@ -114,6 +131,7 @@ def play(env, mode, ngames, max_steps, seeds, savedir, no_render, render_mode, d
                 print(blstats)
 
         action = get_action(env, mode, is_raw_env)
+
         if action is None:
             break
 
@@ -194,7 +212,7 @@ def main():
     parser.add_argument(
         "--max-steps",
         type=int,
-        default=10000,
+        default=1_000_000,
         help="Number of maximum steps per episode.",
     )
     parser.add_argument(
@@ -219,6 +237,12 @@ def main():
         choices=["human", "full", "ansi"],
         help="Render mode. Defaults to 'human'.",
     )
+    parser.add_argument(
+        "--print-frames-separately",
+        "-p",
+        action="store_true",
+        help="Don't overwrite frames, print them all.",
+    )
     flags = parser.parse_args()
 
     if flags.debug: