Remove unittest envs (#2553)

gym/envs/__init__.py

@@ -263,28 +263,3 @@ register(
     entry_point="gym.envs.mujoco:HumanoidStandupEnv",
     max_episode_steps=1000,
 )
-
-# Unit test
-# ---------
-
-register(
-    id="CubeCrash-v0",
-    entry_point="gym.envs.unittest:CubeCrash",
-    reward_threshold=0.9,
-)
-register(
-    id="CubeCrashSparse-v0",
-    entry_point="gym.envs.unittest:CubeCrashSparse",
-    reward_threshold=0.9,
-)
-register(
-    id="CubeCrashScreenBecomesBlack-v0",
-    entry_point="gym.envs.unittest:CubeCrashScreenBecomesBlack",
-    reward_threshold=0.9,
-)
-
-register(
-    id="MemorizeDigits-v0",
-    entry_point="gym.envs.unittest:MemorizeDigits",
-    reward_threshold=20,
-)

gym/envs/unittest/__init__.py

@@ -1,4 +0,0 @@
-from gym.envs.unittest.cube_crash import CubeCrash
-from gym.envs.unittest.cube_crash import CubeCrashSparse
-from gym.envs.unittest.cube_crash import CubeCrashScreenBecomesBlack
-from gym.envs.unittest.memorize_digits import MemorizeDigits

gym/envs/unittest/cube_crash.py

@@ -1,172 +0,0 @@
-from typing import Optional
-
-import numpy as np
-import gym
-from gym import spaces
-from gym.utils import seeding
-
-# Unit test environment for CNNs and CNN+RNN algorithms.
-# Looks like this (RGB observations):
-#
-#  ---------------------------
-# |                           |
-# |                           |
-# |                           |
-# |          **               |
-# |          **               |
-# |                           |
-# |                           |
-# |                           |
-# |                           |
-# |                           |
-#  ========     ==============
-#
-# Goal is to go through the hole at the bottom. Agent controls square using Left-Nop-Right actions.
-# It falls down automatically, episode length is a bit less than FIELD_H
-#
-# CubeCrash-v0                    # shaped reward
-# CubeCrashSparse-v0              # reward 0 or 1 at the end
-# CubeCrashScreenBecomesBlack-v0  # for RNNs
-#
-# To see how it works, run:
-#
-# python examples/agents/keyboard_agent.py CubeCrashScreen-v0
-
-FIELD_W = 32
-FIELD_H = 40
-HOLE_WIDTH = 8
-
-color_black = np.array((0, 0, 0)).astype("float32")
-color_white = np.array((255, 255, 255)).astype("float32")
-color_green = np.array((0, 255, 0)).astype("float32")
-
-
-class CubeCrash(gym.Env):
-    metadata = {
-        "render.modes": ["human", "rgb_array"],
-        "video.frames_per_second": 60,
-        "video.res_w": FIELD_W,
-        "video.res_h": FIELD_H,
-    }
-
-    use_shaped_reward = True
-    use_black_screen = False
-    use_random_colors = False  # Makes env too hard
-
-    def __init__(self):
-        self.viewer = None
-
-        self.observation_space = spaces.Box(
-            0, 255, (FIELD_H, FIELD_W, 3), dtype=np.uint8
-        )
-        self.action_space = spaces.Discrete(3)
-
-        self.reset()
-
-    def random_color(self):
-        return np.array(
-            [
-                self.np_random.integers(low=0, high=255),
-                self.np_random.integers(low=0, high=255),
-                self.np_random.integers(low=0, high=255),
-            ]
-        ).astype("uint8")
-
-    def reset(self, seed: Optional[int] = None):
-        super().reset(seed=seed)
-        self.cube_x = self.np_random.integers(low=3, high=FIELD_W - 3)
-        self.cube_y = self.np_random.integers(low=3, high=FIELD_H // 6)
-        self.hole_x = self.np_random.integers(low=HOLE_WIDTH, high=FIELD_W - HOLE_WIDTH)
-        self.bg_color = self.random_color() if self.use_random_colors else color_black
-        self.potential = None
-        self.step_n = 0
-        while 1:
-            self.wall_color = (
-                self.random_color() if self.use_random_colors else color_white
-            )
-            self.cube_color = (
-                self.random_color() if self.use_random_colors else color_green
-            )
-            if (
-                np.linalg.norm(self.wall_color - self.bg_color) < 50
-                or np.linalg.norm(self.cube_color - self.bg_color) < 50
-            ):
-                continue
-            break
-
-        return self.step(0)[0]
-
-    def step(self, action):
-        if action == 0:
-            pass
-        elif action == 1:
-            self.cube_x -= 1
-        elif action == 2:
-            self.cube_x += 1
-        else:
-            assert 0, "Action %i is out of range" % action
-        self.cube_y += 1
-        self.step_n += 1
-
-        obs = np.zeros((FIELD_H, FIELD_W, 3), dtype=np.uint8)
-        obs[:, :, :] = self.bg_color
-        obs[FIELD_H - 5 : FIELD_H, :, :] = self.wall_color
-        obs[
-            FIELD_H - 5 : FIELD_H,
-            self.hole_x - HOLE_WIDTH // 2 : self.hole_x + HOLE_WIDTH // 2 + 1,
-            :,
-        ] = self.bg_color
-        obs[
-            self.cube_y - 1 : self.cube_y + 2, self.cube_x - 1 : self.cube_x + 2, :
-        ] = self.cube_color
-        if self.use_black_screen and self.step_n > 4:
-            obs[:] = np.zeros((3,), dtype=np.uint8)
-
-        done = False
-        reward = 0
-        dist = np.abs(self.cube_x - self.hole_x)
-        if self.potential is not None and self.use_shaped_reward:
-            reward = (self.potential - dist) * 0.01
-        self.potential = dist
-
-        if self.cube_x - 1 < 0 or self.cube_x + 1 >= FIELD_W:
-            done = True
-            reward = -1
-        elif self.cube_y + 1 >= FIELD_H - 5:
-            if dist >= HOLE_WIDTH // 2:
-                done = True
-                reward = -1
-            elif self.cube_y == FIELD_H:
-                done = True
-                reward = +1
-        self.last_obs = obs
-        return obs, reward, done, {}
-
-    def render(self, mode="human"):
-        if mode == "rgb_array":
-            return self.last_obs
-
-        elif mode == "human":
-            from gym.utils import pyglet_rendering
-
-            if self.viewer is None:
-                self.viewer = pyglet_rendering.SimpleImageViewer()
-            self.viewer.imshow(self.last_obs)
-            return self.viewer.isopen
-
-        else:
-            assert 0, f"Render mode '{mode}' is not supported"
-
-    def close(self):
-        if self.viewer is not None:
-            self.viewer.close()
-            self.viewer = None
-
-
-class CubeCrashSparse(CubeCrash):
-    use_shaped_reward = False
-
-
-class CubeCrashScreenBecomesBlack(CubeCrash):
-    use_shaped_reward = False
-    use_black_screen = True

gym/envs/unittest/memorize_digits.py

@@ -1,144 +0,0 @@
-from typing import Optional
-
-import numpy as np
-import gym
-from gym import spaces
-from gym.utils import seeding
-
-# Unit test environment for CNNs.
-# Looks like this (RGB observations):
-#
-#  ---------------------------
-# |                           |
-# |         ******            |
-# |         ******            |
-# |       **      **          |
-# |       **      **          |
-# |               **          |
-# |               **          |
-# |           ****            |
-# |           ****            |
-# |       ****                |
-# |       ****                |
-# |       **********          |
-# |       **********          |
-# |                           |
-#  ---------------------------
-#
-# Agent should hit action 2 to gain reward. Catches off-by-one errors in your agent.
-#
-# To see how it works, run:
-#
-# python examples/agents/keyboard_agent.py MemorizeDigits-v0
-
-FIELD_W = 32
-FIELD_H = 24
-
-bogus_mnist = [
-    [" **** ", "*    *", "*    *", "*    *", "*    *", " **** "],
-    ["  **  ", " * *  ", "   *  ", "   *  ", "   *  ", "  *** "],
-    [" **** ", "*    *", "     *", "  *** ", "**    ", "******"],
-    [" **** ", "*    *", "   ** ", "     *", "*    *", " **** "],
-    [" *  * ", " *  * ", " *  * ", " **** ", "    * ", "    * "],
-    [" **** ", " *    ", " **** ", "    * ", "    * ", " **** "],
-    ["  *** ", " *    ", " **** ", " *  * ", " *  * ", " **** "],
-    [" **** ", "    * ", "   *  ", "   *  ", "  *   ", "  *   "],
-    [" **** ", "*    *", " **** ", "*    *", "*    *", " **** "],
-    [" **** ", "*    *", "*    *", " *****", "     *", " **** "],
-]
-
-color_black = np.array((0, 0, 0)).astype("float32")
-color_white = np.array((255, 255, 255)).astype("float32")
-
-
-class MemorizeDigits(gym.Env):
-    metadata = {
-        "render.modes": ["human", "rgb_array"],
-        "video.frames_per_second": 60,
-        "video.res_w": FIELD_W,
-        "video.res_h": FIELD_H,
-    }
-
-    use_random_colors = False
-
-    def __init__(self):
-        self.viewer = None
-        self.observation_space = spaces.Box(
-            0, 255, (FIELD_H, FIELD_W, 3), dtype=np.uint8
-        )
-        self.action_space = spaces.Discrete(10)
-        self.bogus_mnist = np.zeros((10, 6, 6), dtype=np.uint8)
-        for digit in range(10):
-            for y in range(6):
-                self.bogus_mnist[digit, y, :] = [
-                    ord(char) for char in bogus_mnist[digit][y]
-                ]
-        self.reset()
-
-    def random_color(self):
-        return np.array(
-            [
-                self.np_random.integers(low=0, high=255),
-                self.np_random.integers(low=0, high=255),
-                self.np_random.integers(low=0, high=255),
-            ]
-        ).astype("uint8")
-
-    def reset(self, seed: Optional[int] = None):
-        super().reset(seed=seed)
-        self.digit_x = self.np_random.integers(low=FIELD_W // 5, high=FIELD_W // 5 * 4)
-        self.digit_y = self.np_random.integers(low=FIELD_H // 5, high=FIELD_H // 5 * 4)
-        self.color_bg = self.random_color() if self.use_random_colors else color_black
-        self.step_n = 0
-        while 1:
-            self.color_digit = (
-                self.random_color() if self.use_random_colors else color_white
-            )
-            if np.linalg.norm(self.color_digit - self.color_bg) < 50:
-                continue
-            break
-        self.digit = -1
-        return self.step(0)[0]
-
-    def step(self, action):
-        reward = -1
-        done = False
-        self.step_n += 1
-        if self.digit == -1:
-            pass
-        else:
-            if self.digit == action:
-                reward = +1
-            done = self.step_n > 20 and 0 == self.np_random.integers(low=0, high=5)
-        self.digit = self.np_random.integers(low=0, high=10)
-        obs = np.zeros((FIELD_H, FIELD_W, 3), dtype=np.uint8)
-        obs[:, :, :] = self.color_bg
-        digit_img = np.zeros((6, 6, 3), dtype=np.uint8)
-        digit_img[:] = self.color_bg
-        xxx = self.bogus_mnist[self.digit] == 42
-        digit_img[xxx] = self.color_digit
-        obs[
-            self.digit_y - 3 : self.digit_y + 3, self.digit_x - 3 : self.digit_x + 3
-        ] = digit_img
-        self.last_obs = obs
-        return obs, reward, done, {}
-
-    def render(self, mode="human"):
-        if mode == "rgb_array":
-            return self.last_obs
-
-        elif mode == "human":
-            from gym.utils import pyglet_rendering
-
-            if self.viewer is None:
-                self.viewer = pyglet_rendering.SimpleImageViewer()
-            self.viewer.imshow(self.last_obs)
-            return self.viewer.isopen
-
-        else:
-            assert 0, f"Render mode '{mode}' is not supported"
-
-    def close(self):
-        if self.viewer is not None:
-            self.viewer.close()
-            self.viewer = None

tests/vector/test_async_vector_env.py

@@ -16,7 +16,7 @@ from gym.vector.async_vector_env import AsyncVectorEnv
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_create_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
     finally:
@@ -27,7 +27,7 @@ def test_create_async_vector_env(shared_memory):
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_reset_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
         observations = env.reset()
@@ -44,7 +44,7 @@ def test_reset_async_vector_env(shared_memory):
 @pytest.mark.parametrize("shared_memory", [True, False])
 @pytest.mark.parametrize("use_single_action_space", [True, False])
 def test_step_async_vector_env(shared_memory, use_single_action_space):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
         observations = env.reset()
@@ -79,24 +79,22 @@ def test_step_async_vector_env(shared_memory, use_single_action_space):
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_copy_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory, copy=True)
         observations = env.reset()
-        observations[0] = 128
+        observations[0] = 0
-        assert not np.all(env.observations[0] == 128)
     finally:
         env.close()
 
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_no_copy_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory, copy=False)
         observations = env.reset()
-        observations[0] = 128
+        observations[0] = 0
-        assert np.all(env.observations[0] == 128)
     finally:
         env.close()
 
@@ -129,7 +127,7 @@ def test_step_timeout_async_vector_env(shared_memory):
 @pytest.mark.filterwarnings("ignore::UserWarning")
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_reset_out_of_order_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(4)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(4)]
     with pytest.raises(NoAsyncCallError):
         try:
             env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
@@ -157,7 +155,7 @@ def test_reset_out_of_order_async_vector_env(shared_memory):
 @pytest.mark.filterwarnings("ignore::UserWarning")
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_step_out_of_order_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(4)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(4)]
     with pytest.raises(NoAsyncCallError):
         try:
             env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
@@ -185,7 +183,7 @@ def test_step_out_of_order_async_vector_env(shared_memory):
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_already_closed_async_vector_env(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(4)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(4)]
     with pytest.raises(ClosedEnvironmentError):
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
         env.close()
@@ -194,10 +192,10 @@ def test_already_closed_async_vector_env(shared_memory):
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_check_spaces_async_vector_env(shared_memory):
-    # CubeCrash-v0 - observation_space: Box(40, 32, 3), action_space: Discrete(3)
+    # CartPole-v1 - observation_space: Box(4,), action_space: Discrete(2)
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
-    # MemorizeDigits-v0 - observation_space: Box(24, 32, 3), action_space: Discrete(10)
+    # FrozenLake-v1 - Discrete(16), action_space: Discrete(4)
-    env_fns[1] = make_env("MemorizeDigits-v0", 1)
+    env_fns[1] = make_env("FrozenLake-v1", 1)
     with pytest.raises(RuntimeError):
         env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)
         env.close(terminate=True)

tests/vector/test_sync_vector_env.py

@@ -8,7 +8,7 @@ from gym.vector.sync_vector_env import SyncVectorEnv
 
 
 def test_create_sync_vector_env():
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("FrozenLake-v1", i) for i in range(8)]
     try:
         env = SyncVectorEnv(env_fns)
     finally:
@@ -18,7 +18,7 @@ def test_create_sync_vector_env():
 
 
 def test_reset_sync_vector_env():
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
     try:
         env = SyncVectorEnv(env_fns)
         observations = env.reset()
@@ -34,7 +34,7 @@ def test_reset_sync_vector_env():
 
 @pytest.mark.parametrize("use_single_action_space", [True, False])
 def test_step_sync_vector_env(use_single_action_space):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("FrozenLake-v1", i) for i in range(8)]
     try:
         env = SyncVectorEnv(env_fns)
         observations = env.reset()
@@ -50,7 +50,7 @@ def test_step_sync_vector_env(use_single_action_space):
     finally:
         env.close()
 
-    assert isinstance(env.observation_space, Box)
+    assert isinstance(env.observation_space, MultiDiscrete)
     assert isinstance(observations, np.ndarray)
     assert observations.dtype == env.observation_space.dtype
     assert observations.shape == (8,) + env.single_observation_space.shape
@@ -68,10 +68,10 @@ def test_step_sync_vector_env(use_single_action_space):
 
 
 def test_check_spaces_sync_vector_env():
-    # CubeCrash-v0 - observation_space: Box(40, 32, 3), action_space: Discrete(3)
+    # CartPole-v1 - observation_space: Box(4,), action_space: Discrete(2)
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(8)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(8)]
-    # MemorizeDigits-v0 - observation_space: Box(24, 32, 3), action_space: Discrete(10)
+    # FrozenLake-v1 - Discrete(16), action_space: Discrete(4)
-    env_fns[1] = make_env("MemorizeDigits-v0", 1)
+    env_fns[1] = make_env("FrozenLake-v1", 1)
     with pytest.raises(RuntimeError):
         env = SyncVectorEnv(env_fns)
         env.close()

tests/vector/test_vector_env.py

@@ -11,7 +11,7 @@ from gym.vector.vector_env import VectorEnv
 
 @pytest.mark.parametrize("shared_memory", [True, False])
 def test_vector_env_equal(shared_memory):
-    env_fns = [make_env("CubeCrash-v0", i) for i in range(4)]
+    env_fns = [make_env("CartPole-v1", i) for i in range(4)]
     num_steps = 100
     try:
         async_env = AsyncVectorEnv(env_fns, shared_memory=shared_memory)