Boxohack fix

.pre-commit-config.yaml

@@ -1,6 +1,6 @@
 repos:
 -   repo: https://github.com/ambv/black
-    rev: stable
+    rev: 23.12.1
     hooks:
     - id: black
       language_version: python3.8

minihack/envs/boxohack.py

@@ -6,9 +6,12 @@
 from importlib_resources import files
 from nle import nethack
 from minihack.envs import register
-from minihack import LevelGenerator, MiniHackNavigation
+from minihack.level_generator import LevelGenerator
+from minihack.navigation import MiniHackNavigation
 
-LEVELS_PATH = files("minihack.dat").joinpath("boxoban-levels-master")
+LEVELS_PATH = (
+    files("minihack").joinpath("dat").joinpath("boxoban-levels-master")
+)
 # The agent can only move towards 4 cardinal directions (instead of default 8)
 MOVE_ACTIONS = tuple(nethack.CompassCardinalDirection)
 
@@ -43,11 +46,11 @@ def __init__(self, *args, **kwargs):
         self._flags = tuple(kwargs.pop("flags", []))
         try:
             self._levels = load_boxoban_levels(cur_levels_path)
-        except FileNotFoundError:
+        except FileNotFoundError as e:
             raise ModuleNotFoundError(
                 "To use Boxoban environments, please download maps using "
                 "the minihack/scripts/download_boxoban_levels.py script."
-            )
+            ) from e
 
         self._reward_shaping_coefficient = kwargs.pop(
             "reward_shaping_coefficient", 0
@@ -56,6 +59,7 @@ def __init__(self, *args, **kwargs):
         super().__init__(
             *args, des_file=self.get_lvl_gen().get_des(), **kwargs
         )
+        self._goal_pos_set = None
 
     def get_env_map(self, level):
         info = {"fountains": [], "boulders": []}
@@ -90,14 +94,18 @@ def get_lvl_gen(self):
         lvl_gen.set_start_pos(info["player"])
         return lvl_gen
 
-    def reset(self, options=dict(wizkit_items=None)):
+    def reset(self, *args, **kwargs):
         self.update(self.get_lvl_gen().get_des())
-        initial_obs = super().reset(options=options)
+        if "options" not in kwargs:
+            kwargs["options"] = dict(wizkit_items=None)
+        initial_obs = super().reset(*args, **kwargs)
         self._goal_pos_set = self._object_positions(self.last_observation, "{")
         return initial_obs
 
     def _is_episode_end(self, observation):
         # If every boulder is on a fountain, we're done
+        if self._goal_pos_set is None:
+            return self.StepStatus.RUNNING
         if self._goal_pos_set == self._object_positions(observation, "`"):
             return self.StepStatus.TASK_SUCCESSFUL
         else:

minihack/scripts/download_boxoban_levels.py

@@ -4,7 +4,7 @@
 import zipfile
 from importlib_resources import files
 
-DESTINATION_PATH = files("minihack.dat")
+DESTINATION_PATH = files("minihack").joinpath("dat")
 BOXOBAN_REPO_URL = (
     "https://github.com/deepmind/boxoban-levels/archive/refs/heads/master.zip"
 )
@@ -14,7 +14,7 @@ def download_boxoban_levels():
     print("Downloading Boxoban levels...")
     os.system(
         f"wget -c --read-timeout=5 --tries=0 "
-        f'"{BOXOBAN_REPO_URL}" -P {DESTINATION_PATH}'
+        f"{BOXOBAN_REPO_URL!r} -P {DESTINATION_PATH}"
     )
     print("Boxoban levels downloaded, unpacking...")
 

minihack/tests/test_envs.py

@@ -6,6 +6,7 @@
 import sys
 import tempfile
 
+import shutil
 import gymnasium as gym
 import numpy as np
 import pytest
@@ -14,6 +15,9 @@
 import nle
 from nle import nethack
 
+from minihack.scripts.download_boxoban_levels import download_boxoban_levels
+from minihack.envs.boxohack import LEVELS_PATH
+
 
 def get_minihack_env_ids():
     specs = gym.envs.registry.keys()
@@ -163,7 +167,8 @@ def test_rollout(self, env_name, rollout_len):
             assert os.path.exists(
                 os.path.join(
                     savedir,
-                    "nle.%i.0.ttyrec%i.bz2" % (os.getpid(), nethack.TTYREC_VERSION),
+                    "nle.%i.0.ttyrec%i.bz2"
+                    % (os.getpid(), nethack.TTYREC_VERSION),
                 )
             )
             assert os.path.exists(
@@ -290,12 +295,16 @@ def test_reward(self, env):
         _, _ = env.reset()
 
         for _ in range(4):
-            _, reward, done, _, _ = env.step(env.unwrapped.actions.index(ord("j")))
+            _, reward, done, _, _ = env.step(
+                env.unwrapped.actions.index(ord("j"))
+            )
             assert reward == 0.0
             assert not done
 
         for _ in range(3):
-            _, reward, done, _, _ = env.step(env.unwrapped.actions.index(ord("l")))
+            _, reward, done, _, _ = env.step(
+                env.unwrapped.actions.index(ord("l"))
+            )
             assert reward == 0.0
             assert not done
 
@@ -304,3 +313,32 @@ def test_reward(self, env):
 
         assert done
         assert reward == 1.0
+
+
+@pytest.mark.parametrize(
+    "env_name",
+    [
+        "MiniHack-Boxoban-Unfiltered-v0",
+        "MiniHack-Boxoban-Medium-v0",
+        "MiniHack-Boxoban-Hard-v0",
+    ],
+)
+class TestBoxoban:
+    @pytest.fixture(autouse=True)
+    def setup_boxoban_levels(self):
+        self.levels_downloaded = False
+        if not os.path.isdir(LEVELS_PATH):
+            download_boxoban_levels()
+            self.levels_downloaded = True
+
+        yield
+
+        if self.levels_downloaded and os.path.isdir(LEVELS_PATH):
+            shutil.rmtree(LEVELS_PATH)
+
+    def test_boxoban_init(self, env_name):
+        env = gym.make(env_name)
+        env.reset()
+        for _ in range(5):
+            env.step(env.action_space.sample())
+        env.close()