Add compatibility with numpy 2.0 (#1094)

Co-authored-by: Mark Towers <mark.m.towers@gmail.com>

.github/workflows/build.yml

@@ -10,15 +10,21 @@ jobs:
     strategy:
       matrix:
         python-version: ['3.8', '3.9', '3.10', '3.11', '3.12']
+        numpy-version: ['>=1.21,<2.0', '>=2.0']
+        exclude:
+          - python-version: '3.8'  # numpy>=2.0 requires Python>=3.9
+            numpy-version: '>=2.0'
     steps:
       - uses: actions/checkout@v4
       - run: |
-           docker build -f bin/all-py.Dockerfile \
-             --build-arg PYTHON_VERSION=${{ matrix.python-version }} \
-             --tag gymnasium-all-docker .
+          docker build -f bin/all-py.Dockerfile \
+            --build-arg PYTHON_VERSION="${{ matrix.python-version }}" \
+            --build-arg NUMPY_VERSION="${{ matrix.numpy-version }}" \
+            --tag gymnasium-all-docker .
       - name: Run tests
         run: docker run gymnasium-all-docker pytest tests/*
       - name: Run doctests
+        if: ${{ matrix.numpy-version == '>=2.0' }}
         run: docker run gymnasium-all-docker pytest --doctest-modules gymnasium/
 
   build-necessary:

bin/all-py.Dockerfile

@@ -1,5 +1,6 @@
 # A Dockerfile that sets up a full Gymnasium install with test dependencies
 ARG PYTHON_VERSION
+ARG NUMPY_VERSION=">=1.21,<2.0"
 FROM python:$PYTHON_VERSION
 
 SHELL ["/bin/bash", "-o", "pipefail", "-c"]
@@ -27,6 +28,9 @@ RUN git clone https://github.com/openai/mujoco-py.git\
 COPY . /usr/local/gymnasium/
 WORKDIR /usr/local/gymnasium/
 
+# Specify the numpy version to cover both 1.x and 2.x
+RUN pip install --upgrade "numpy$NUMPY_VERSION"
+
 # Test with PyTorch CPU build, since CUDA is not available in CI anyway
 RUN pip install .[all,testing] --no-cache-dir --extra-index-url https://download.pytorch.org/whl/cpu
 

gymnasium/envs/box2d/car_racing.py

@@ -540,6 +540,7 @@ def reset(
     def step(self, action: Union[np.ndarray, int]):
         assert self.car is not None
         if action is not None:
+            action = action.astype(np.float64)
             if self.continuous:
                 self.car.steer(-action[0])
                 self.car.gas(action[1])

gymnasium/envs/box2d/lunar_lander.py

@@ -509,7 +509,7 @@ def step(self, action):
             )
 
         if self.continuous:
-            action = np.clip(action, -1, +1).astype(np.float32)
+            action = np.clip(action, -1, +1).astype(np.float64)
         else:
             assert self.action_space.contains(
                 action

gymnasium/envs/classic_control/acrobot.py

@@ -447,9 +447,9 @@ def rk4(derivs, y0, t):
     try:
         Ny = len(y0)
     except TypeError:
-        yout = np.zeros((len(t),), np.float_)
+        yout = np.zeros((len(t),), np.float64)
     else:
-        yout = np.zeros((len(t), Ny), np.float_)
+        yout = np.zeros((len(t), Ny), np.float64)
 
     yout[0] = y0
 

gymnasium/envs/classic_control/pendulum.py

@@ -241,7 +241,10 @@ def render(self):
         if self.last_u is not None:
             scale_img = pygame.transform.smoothscale(
                 img,
-                (scale * np.abs(self.last_u) / 2, scale * np.abs(self.last_u) / 2),
+                (
+                    float(scale * np.abs(self.last_u) / 2),
+                    float(scale * np.abs(self.last_u) / 2),
+                ),
             )
             is_flip = bool(self.last_u > 0)
             scale_img = pygame.transform.flip(scale_img, is_flip, True)

gymnasium/spaces/dict.py

@@ -20,7 +20,7 @@ class Dict(Space[typing.Dict[str, Any]], typing.Mapping[str, Space[Any]]):
         >>> from gymnasium.spaces import Dict, Box, Discrete
         >>> observation_space = Dict({"position": Box(-1, 1, shape=(2,)), "color": Discrete(3)}, seed=42)
         >>> observation_space.sample()
-        {'color': 0, 'position': array([-0.3991573 ,  0.21649833], dtype=float32)}
+        {'color': np.int64(0), 'position': array([-0.3991573 ,  0.21649833], dtype=float32)}
 
         With a nested dict:
 

gymnasium/spaces/discrete.py

@@ -18,10 +18,10 @@ class Discrete(Space[np.int64]):
         >>> from gymnasium.spaces import Discrete
         >>> observation_space = Discrete(2, seed=42) # {0, 1}
         >>> observation_space.sample()
-        0
+        np.int64(0)
         >>> observation_space = Discrete(3, start=-1, seed=42)  # {-1, 0, 1}
         >>> observation_space.sample()
-        -1
+        np.int64(-1)
     """
 
     def __init__(

gymnasium/spaces/oneof.py

@@ -19,9 +19,9 @@ class OneOf(Space[Any]):
         >>> from gymnasium.spaces import OneOf, Box, Discrete
         >>> observation_space = OneOf((Discrete(2), Box(-1, 1, shape=(2,))), seed=123)
         >>> observation_space.sample()  # the first element is the space index (Box in this case) and the second element is the sample from Box
-        (0, 0)
+        (np.int64(0), np.int64(0))
         >>> observation_space.sample()  # this time the Discrete space was sampled as index=0
-        (1, array([-0.00711833, -0.7257502 ], dtype=float32))
+        (np.int64(1), array([-0.00711833, -0.7257502 ], dtype=float32))
         >>> observation_space[0]
         Discrete(2)
         >>> observation_space[1]

gymnasium/spaces/tuple.py

@@ -19,7 +19,7 @@ class Tuple(Space[typing.Tuple[Any, ...]], typing.Sequence[Any]):
         >>> from gymnasium.spaces import Tuple, Box, Discrete
         >>> observation_space = Tuple((Discrete(2), Box(-1, 1, shape=(2,))), seed=42)
         >>> observation_space.sample()
-        (0, array([-0.3991573 ,  0.21649833], dtype=float32))
+        (np.int64(0), array([-0.3991573 ,  0.21649833], dtype=float32))
     """
 
     def __init__(

gymnasium/utils/passive_env_checker.py

@@ -32,15 +32,9 @@ def _check_box_observation_space(observation_space: spaces.Box):
     ), f"The Box observation space shape and high shape have have different shapes, high shape: {observation_space.high.shape}, box shape: {observation_space.shape}"
 
     if np.any(observation_space.low == observation_space.high):
-        logger.warn(
-            "A Box observation space maximum and minimum values are equal. "
-            f"Actual equal coordinates: {[x for x in zip(*np.where(observation_space.low == observation_space.high))]}"
-        )
+        logger.warn("A Box observation space maximum and minimum values are equal.")
     elif np.any(observation_space.high < observation_space.low):
-        logger.warn(
-            "A Box observation space low value is greater than a high value. "
-            f"Actual less than coordinates: {[x for x in zip(*np.where(observation_space.high < observation_space.low))]}"
-        )
+        logger.warn("A Box observation space low value is greater than a high value.")
 
 
 def _check_box_action_space(action_space: spaces.Box):
@@ -57,10 +51,7 @@ def _check_box_action_space(action_space: spaces.Box):
     ), f"The Box action space shape and high shape have different shapes, high shape: {action_space.high.shape}, box shape: {action_space.shape}"
 
     if np.any(action_space.low == action_space.high):
-        logger.warn(
-            "A Box action space maximum and minimum values are equal. "
-            f"Actual equal coordinates: {[x for x in zip(*np.where(action_space.low == action_space.high))]}"
-        )
+        logger.warn("A Box action space maximum and minimum values are equal.")
 
 
 def check_space(

gymnasium/wrappers/atari_preprocessing.py

@@ -37,8 +37,14 @@ class AtariPreprocessing(gym.Wrapper, gym.utils.RecordConstructorArgs):
 
     Example:
         >>> import gymnasium as gym # doctest: +SKIP
-        >>> env = gym.make("ALE/Adventure-v5") # doctest: +SKIP
-        >>> env = AtariPreprocessing(env, noop_max=10, frame_skip=0, screen_size=84, terminal_on_life_loss=True, grayscale_obs=False, grayscale_newaxis=False) # doctest: +SKIP
+        >>> import ale_py # doctest: +SKIP
+        >>> gym.register_envs(ale_py) # doctest: +SKIP
+        >>> env = gym.make("ALE/Pong-v5", frameskip=1) # doctest: +SKIP
+        >>> env = AtariPreprocessing( # doctest: +SKIP
+        ...    env,
+        ...    noop_max=10, frame_skip=4, terminal_on_life_loss=True,
+        ...    screen_size=84, grayscale_obs=False, grayscale_newaxis=False
+        ... )
 
     Change logs:
      * Added in gym v0.12.2 (gym #1455)

gymnasium/wrappers/stateful_observation.py

@@ -559,9 +559,9 @@ class MaxAndSkipObservation(
         >>> wrapped_obs0, *_ = wrapped_env.reset(seed=123)
         >>> wrapped_obs1, *_ = wrapped_env.step(1)
         >>> np.all(obs0 == wrapped_obs0)
-        True
+        np.True_
         >>> np.all(wrapped_obs1 == skip_and_max_obs)
-        True
+        np.True_
 
     Change logs:
      * v1.0.0 - Initially add

gymnasium/wrappers/stateful_reward.py

@@ -52,7 +52,7 @@ class NormalizeReward(
         ...
         >>> env.close()
         >>> np.var(episode_rewards)
-        0.0008876301247721108
+        np.float64(0.0008876301247721108)
 
     Example with the normalize reward wrapper:
         >>> import numpy as np
@@ -70,7 +70,7 @@ class NormalizeReward(
         >>> env.close()
         >>> # will approach 0.99 with more episodes
         >>> np.var(episode_rewards)
-        0.010162116476634746
+        np.float64(0.010162116476634746)
 
     Change logs:
      * v0.21.0 - Initially added

gymnasium/wrappers/transform_action.py

@@ -144,7 +144,7 @@ class RescaleAction(
         >>> wrapped_env = RescaleAction(env, min_action=min_action, max_action=max_action)
         >>> wrapped_env_obs, _, _, _, _ = wrapped_env.step(max_action)
         >>> np.all(obs == wrapped_env_obs)
-        True
+        np.True_
 
     Change logs:
      * v0.15.4 - Initially added

gymnasium/wrappers/transform_observation.py

@@ -632,11 +632,11 @@ class AddRenderObservation(
         >>> obs, _ = env.reset(seed=123)
         >>> image = env.render()
         >>> np.all(obs == image)
-        True
+        np.True_
         >>> obs, *_ = env.step(env.action_space.sample())
         >>> image = env.render()
         >>> np.all(obs == image)
-        True
+        np.True_
 
     Example - Add the rendered image to the original observation as a dictionary item:
         >>> env = gym.make("CartPole-v1", render_mode="rgb_array")
@@ -649,11 +649,11 @@ class AddRenderObservation(
         >>> obs["state"]
         array([ 0.01823519, -0.0446179 , -0.02796401, -0.03156282], dtype=float32)
         >>> np.all(obs["pixels"] == env.render())
-        True
+        np.True_
         >>> obs, reward, terminates, truncates, info = env.step(env.action_space.sample())
         >>> image = env.render()
         >>> np.all(obs["pixels"] == image)
-        True
+        np.True_
 
     Change logs:
      * v0.15.0 - Initially added as ``PixelObservationWrapper``

gymnasium/wrappers/transform_reward.py

@@ -77,7 +77,7 @@ class ClipReward(TransformReward[ObsType, ActType], gym.utils.RecordConstructorA
         >>> _ = env.reset()
         >>> _, rew, _, _, _ = env.step(1)
         >>> rew
-        0.5
+        np.float64(0.5)
 
     Change logs:
      * v1.0.0 - Initially added

gymnasium/wrappers/vector/dict_info_to_list.py

@@ -50,18 +50,18 @@ class DictInfoToList(VectorWrapper):
     Another example for vector environments:
         >>> import numpy as np
         >>> import gymnasium as gym
-        >>> envs = gym.make_vec("HalfCheetah-v4", num_envs=3)
+        >>> envs = gym.make_vec("HalfCheetah-v4", num_envs=2)
         >>> _ = envs.reset(seed=123)
         >>> _ = envs.action_space.seed(123)
         >>> _, _, _, _, infos = envs.step(envs.action_space.sample())
         >>> infos
-        {'x_position': array([0.03332211, 0.10172355, 0.08920531]), '_x_position': array([ True,  True,  True]), 'x_velocity': array([-0.06296527,  0.89345848,  0.37710836]), '_x_velocity': array([ True,  True,  True]), 'reward_run': array([-0.06296527,  0.89345848,  0.37710836]), '_reward_run': array([ True,  True,  True]), 'reward_ctrl': array([-0.24503503, -0.21944423, -0.20672209]), '_reward_ctrl': array([ True,  True,  True])}
+        {'x_position': array([0.03332211, 0.10172355]), '_x_position': array([ True,  True]), 'x_velocity': array([-0.06296527,  0.89345848]), '_x_velocity': array([ True,  True]), 'reward_run': array([-0.06296527,  0.89345848]), '_reward_run': array([ True,  True]), 'reward_ctrl': array([-0.24503504, -0.21944423], dtype=float32), '_reward_ctrl': array([ True,  True])}
         >>> envs = DictInfoToList(envs)
         >>> _ = envs.reset(seed=123)
         >>> _ = envs.action_space.seed(123)
         >>> _, _, _, _, infos = envs.step(envs.action_space.sample())
         >>> infos
-        [{'x_position': 0.03332210900362942, 'x_velocity': -0.06296527291998533, 'reward_run': -0.06296527291998533, 'reward_ctrl': -0.2450350284576416}, {'x_position': 0.10172354684460168, 'x_velocity': 0.8934584807363618, 'reward_run': 0.8934584807363618, 'reward_ctrl': -0.21944422721862794}, {'x_position': 0.08920531470057845, 'x_velocity': 0.3771083596080768, 'reward_run': 0.3771083596080768, 'reward_ctrl': -0.20672209262847902}]
+        [{'x_position': np.float64(0.03332210900362942), 'x_velocity': np.float64(-0.06296527291998533), 'reward_run': np.float64(-0.06296527291998533), 'reward_ctrl': np.float32(-0.24503504)}, {'x_position': np.float64(0.10172354684460168), 'x_velocity': np.float64(0.8934584807363618), 'reward_run': np.float64(0.8934584807363618), 'reward_ctrl': np.float32(-0.21944423)}]
 
     Change logs:
      * v0.24.0 - Initially added as ``VectorListInfo``

gymnasium/wrappers/vector/stateful_observation.py

@@ -35,9 +35,9 @@ class NormalizeObservation(VectorObservationWrapper, gym.utils.RecordConstructor
         >>> for _ in range(100):
         ...     obs, *_ = envs.step(envs.action_space.sample())
         >>> np.mean(obs)
-        0.024251968
+        np.float32(0.024251968)
         >>> np.std(obs)
-        0.62259156
+        np.float32(0.62259156)
         >>> envs.close()
 
     Example with the normalize reward wrapper:
@@ -49,9 +49,9 @@ class NormalizeObservation(VectorObservationWrapper, gym.utils.RecordConstructor
         >>> for _ in range(100):
         ...     obs, *_ = envs.step(envs.action_space.sample())
         >>> np.mean(obs)
-        -0.2359734
+        np.float32(-0.2359734)
         >>> np.std(obs)
-        1.1938739
+        np.float32(1.1938739)
         >>> envs.close()
     """
 

gymnasium/wrappers/vector/stateful_reward.py

@@ -44,9 +44,9 @@ class NormalizeReward(VectorWrapper, gym.utils.RecordConstructorArgs):
         ...
         >>> envs.close()
         >>> np.mean(episode_rewards)
-        -0.03359492141887935
+        np.float64(-0.03359492141887935)
         >>> np.std(episode_rewards)
-        0.029028230434438706
+        np.float64(0.029028230434438706)
 
     Example with the normalize reward wrapper:
         >>> import gymnasium as gym
@@ -62,9 +62,9 @@ class NormalizeReward(VectorWrapper, gym.utils.RecordConstructorArgs):
         ...
         >>> envs.close()
         >>> np.mean(episode_rewards)
-        -0.1598639586606745
+        np.float64(-0.1598639586606745)
         >>> np.std(episode_rewards)
-        0.27800309628058434
+        np.float64(0.27800309628058434)
     """
 
     def __init__(

gymnasium/wrappers/vector/vectorize_action.py

@@ -33,7 +33,7 @@ class TransformAction(VectorActionWrapper):
         >>> obs
         array([[-0.46553135, -0.00142543],
                [-0.498371  , -0.00715587],
-               [-0.4651575 , -0.00624371]], dtype=float32)
+               [-0.46515748, -0.00624371]], dtype=float32)
 
     Example - With action transformation:
         >>> import gymnasium as gym

gymnasium/wrappers/vector/vectorize_observation.py

@@ -321,15 +321,15 @@ class RescaleObservation(VectorizeTransformObservation):
         >>> envs = gym.make_vec("CartPole-v1", num_envs=3, vectorization_mode="sync")
         >>> obs, info = envs.reset(seed=123)
         >>> obs.min()
-        -0.0446179
+        np.float32(-0.0446179)
         >>> obs.max()
-        0.0469136
+        np.float32(0.0469136)
         >>> envs = RescaleObservation(envs, min_obs=-5.0, max_obs=5.0)
         >>> obs, info = envs.reset(seed=123)
         >>> obs.min()
-        -0.33379582
+        np.float32(-0.33379582)
         >>> obs.max()
-        0.55998987
+        np.float32(0.55998987)
         >>> envs.close()
     """
 

pyproject.toml

@@ -108,7 +108,6 @@ gymnasium = [
 # Linters and Test tools #######################################################
 
 [tool.black]
-safe = true
 
 [tool.isort]
 atomic = true

tests/envs/test_action_dim_check.py

@@ -127,19 +127,17 @@ def test_box_actions_out_of_bound(env: gym.Env):
         if is_upper_bound:
             obs, _, _, _, _ = env.step(upper_bounds)
             oob_action = upper_bounds.copy()
-            oob_action[i] += np.cast[dtype](OOB_VALUE)
+            oob_action[i] += np.asarray(OOB_VALUE, dtype=dtype)
 
             assert oob_action[i] > upper_bounds[i]
             oob_obs, _, _, _, _ = oob_env.step(oob_action)
 
             assert np.all(obs == oob_obs)
 
         if is_lower_bound:
-            obs, _, _, _, _ = env.step(
-                lower_bounds
-            )  # `env` is unwrapped, and in new step API
+            obs, _, _, _, _ = env.step(lower_bounds)
             oob_action = lower_bounds.copy()
-            oob_action[i] -= np.cast[dtype](OOB_VALUE)
+            oob_action[i] -= np.asarray(OOB_VALUE, dtype=dtype)
 
             assert oob_action[i] < lower_bounds[i]
             oob_obs, _, _, _, _ = oob_env.step(oob_action)

tests/spaces/test_spaces.py

@@ -239,7 +239,7 @@ def chi2_test(sample, low, high, bounded_below, bounded_above):
 
     if bounded_below and bounded_above:
         # X ~ U(low, high)
-        degrees_of_freedom = high - low + 1
+        degrees_of_freedom = int(high) - int(low) + 1
         observed_frequency = np.bincount(sample - low, minlength=degrees_of_freedom)
         assert observed_frequency.shape == (degrees_of_freedom,)
         expected_frequency = np.ones(degrees_of_freedom) * n_trials / degrees_of_freedom