Revert "Revert "Revert "[RLlib] Upgrade to gymnasium 1.0.0 (ale_py 0.…

…10.1, mujoco 3.2.4, pettingzoo 1.24.3 supersuit 3.9.3).""" (ray-project#48435)

Reverts ray-project#48308

Signed-off-by: mohitjain2504 <mohit.jain@dream11.com>

doc/source/ray-core/examples/plot_pong_example.ipynb

@@ -292,7 +292,7 @@
     "@ray.remote\n",
     "class RolloutWorker(object):\n",
     "    def __init__(self):\n",
-    "        self.env = gym.make(\"ale_py:ALE/Pong-v5\")\n",
+    "        self.env = gym.make(\"ALE/Pong-v5\")\n",
     "\n",
     "    def compute_gradient(self, model):\n",
     "        # Compute a simulation episode.\n",

doc/source/rllib/doc_code/dreamerv3_inference.py

@@ -10,7 +10,7 @@
 
 env_name = "CartPole-v1"
 # Use the vector env API.
-env = gym.make_vec(env_name, num_envs=1, vectorization_mode="sync")
+env = gym.vector.make(env_name, num_envs=1, asynchronous=False)
 
 terminated = truncated = False
 # Reset the env.

doc/source/rllib/doc_code/training.py

@@ -4,7 +4,7 @@
 try:
     import gymnasium as gym
 
-    env = gym.make("ale_py:ALE/Pong-v5")
+    env = gym.make("ALE/Pong-v5")
     obs, infos = env.reset()
 except Exception:
     import gym

doc/source/rllib/rllib-examples.rst

@@ -202,7 +202,7 @@ in roughly 5min. It can be run like this on a single g5.24xlarge (or g6.24xlarge
 .. code-block:: bash
 
     $ cd ray/rllib/tuned_examples/ppo
-    $ python atari_ppo.py --env=ale_py:ALE/Pong-v5 --num-gpus=4 --num-env-runners=95
+    $ python atari_ppo.py --env ALE/Pong-v5 --num-gpus=4 --num-env-runners=95
 
 Note that some of the files in this folder are used for RLlib's daily or weekly
 release tests as well.

python/requirements.txt

@@ -38,7 +38,7 @@ colorful
 rich
 opentelemetry-sdk
 fastapi
-gymnasium==1.0.0
+gymnasium==0.28.1
 virtualenv!=20.21.1,>=20.0.24
 opentelemetry-api
 opencensus

python/requirements/ml/rllib-test-requirements.txt

@@ -3,28 +3,43 @@
 # Environment adapters.
 # ---------------------
 # Atari
-imageio==2.34.2
-ale_py==0.10.1
+gymnasium==0.28.1; python_version < "3.12"
+imageio; python_version < "3.12"
+ale_py==0.8.1; python_version < "3.12"
 # For testing MuJoCo envs with gymnasium.
-mujoco==3.2.4
+mujoco==2.3.6; python_version < "3.12"
 dm_control==1.0.12; python_version < "3.12"
 
 # For tests on PettingZoo's multi-agent envs.
-pettingzoo==1.24.3
+pettingzoo==1.23.1
+# When installing pettingzoo, chess is missing, even though its a dependancy
+# TODO: remove if a future pettingzoo and/or ray version fixes this dependancy issue.
+chess==1.7.0
 pymunk==6.2.1
-tinyscaler==1.2.8
-shimmy==2.0.0
-supersuit==3.9.3
+supersuit==3.8.0; python_version < "3.12"
+tinyscaler==1.2.6; python_version < "3.12"
+shimmy
+
+# Kaggle envs.
+kaggle_environments==1.7.11
+# Unity3D testing
+# TODO(sven): Add this back to rllib-requirements.txt once mlagents no longer pins torch<1.9.0 version.
+#mlagents==0.28.0
+mlagents_envs==0.28.0
 
 # For tests on minigrid.
-minigrid==2.3.1
+minigrid
+# For tests on RecSim and Kaggle envs.
+# Explicitly depends on `tensorflow` and doesn't accept `tensorflow-macos`
+recsim==0.2.4; (sys_platform != 'darwin' or platform_machine != 'arm64') and python_version < "3.12"
+# recsim depends on dopamine-rl, but dopamine-rl pins gym <= 0.25.2, which break some envs
+dopamine-rl==4.0.5; (sys_platform != 'darwin' or platform_machine != 'arm64') and python_version < "3.12"
 tensorflow_estimator
 # DeepMind's OpenSpiel
 open-spiel==1.4
-# Unity3D testing
-mlagents_envs==0.28.0
 
 # Requires libtorrent which is unavailable for arm64
+autorom[accept-rom-license]; platform_machine != "arm64"
 h5py==3.10.0
 
 # Requirements for rendering.

python/requirements_compiled.txt

@@ -75,10 +75,10 @@ aiosqlite==0.19.0
     # via ypy-websocket
 alabaster==0.7.13
     # via sphinx
-ale-py==0.10.1
+ale-py==0.8.1 ; python_version < "3.12"
     # via
     #   -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
-    #   gymnasium
+    #   gym
 alembic==1.12.1
     # via
     #   aim
@@ -272,6 +272,8 @@ charset-normalizer==3.3.2
     # via
     #   requests
     #   snowflake-connector-python
+chess==1.7.0
+    # via -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
 chex==0.1.7
     # via optax
 clang-format==12.0.1
@@ -304,6 +306,7 @@ cloudpickle==2.2.0
     #   -r /ray/ci/../python/requirements/test-requirements.txt
     #   dask
     #   distributed
+    #   gym
     #   gymnasium
     #   hyperopt
     #   mlagents-envs
@@ -701,7 +704,13 @@ gsutil==5.27
     # via -r /ray/ci/../python/requirements/docker/ray-docker-requirements.txt
 gunicorn==20.1.0
     # via mlflow
-gymnasium==1.0.0
+gym==0.26.2
+    # via
+    #   dopamine-rl
+    #   recsim
+gym-notices==0.0.8
+    # via gym
+gymnasium==0.28.1 ; python_version < "3.12"
     # via
     #   -r /ray/ci/../python/requirements.txt
     #   -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
@@ -1117,7 +1126,7 @@ msrestazure==0.6.4
     # via
     #   -r /ray/ci/../python/requirements/test-requirements.txt
     #   azure-cli-core
-mujoco==3.2.4
+mujoco==2.3.6 ; python_version < "3.12"
     # via
     #   -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
     #   dm-control
@@ -1237,6 +1246,7 @@ numpy==1.26.4
     #   flax
     #   gpy
     #   gradio
+    #   gym
     #   gymnasium
     #   h5py
     #   hpbandster
@@ -1280,6 +1290,7 @@ numpy==1.26.4
     #   pyro-ppl
     #   pytorch-lightning
     #   raydp
+    #   recsim
     #   scikit-image
     #   scikit-learn
     #   scipy
@@ -1478,7 +1489,7 @@ pbr==6.0.0
     #   sarif-om
 peewee==3.17.0
     # via semgrep
-pettingzoo==1.24.3
+pettingzoo==1.23.1
     # via -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
 pexpect==4.8.0
     # via
@@ -1851,6 +1862,8 @@ querystring-parser==1.2.4
     # via raydp
 raydp==1.7.0b20231020.dev0
     # via -r /ray/ci/../python/requirements/ml/data-test-requirements.txt
+recsim==0.2.4 ; (sys_platform != "darwin" or platform_machine != "arm64") and python_version < "3.12"
+    # via -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
 redis==4.4.2
     # via -r /ray/ci/../python/requirements/test-requirements.txt
 regex==2024.5.15
@@ -2036,7 +2049,7 @@ shellcheck-py==0.7.1.1
     # via -r /ray/ci/../python/requirements/lint-requirements.txt
 shellingham==1.5.4
     # via typer
-shimmy==2.0.0
+shimmy==1.3.0
     # via -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
 shortuuid==1.0.1
     # via -r /ray/ci/../python/requirements/ml/tune-test-requirements.txt
@@ -2154,7 +2167,9 @@ statsmodels==0.14.0
     # via
     #   hpbandster
     #   statsforecast
-supersuit==3.9.3
+strictyaml==1.7.3
+    # via pyiceberg
+supersuit==3.8.0 ; python_version < "3.12"
     # via -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
 sympy==1.13.1
     # via
@@ -2241,7 +2256,7 @@ timm==0.9.2
     # via -r /ray/ci/../python/requirements/ml/tune-test-requirements.txt
 tinycss2==1.3.0
     # via nbconvert
-tinyscaler==1.2.8
+tinyscaler==1.2.6 ; python_version < "3.12"
     # via
     #   -r /ray/ci/../python/requirements/ml/rllib-test-requirements.txt
     #   supersuit

python/setup.py

@@ -299,7 +299,7 @@ def get_packages(self):
 
     setup_spec.extras["rllib"] = setup_spec.extras["tune"] + [
         "dm_tree",
-        "gymnasium==1.0.0",
+        "gymnasium==0.28.1",
         "lz4",
         "scikit-image",
         "pyyaml",

release/long_running_tests/workloads/apex.py

@@ -39,7 +39,7 @@
     {
         "apex": {
             "run": "APEX",
-            "env": "ale_py:ALE/Pong-v5",
+            "env": "ALE/Pong-v5",
             "config": {
                 "num_workers": 3,
                 "num_gpus": 0,

release/ml_user_tests/tune_rllib/run_connect_tests.py

@@ -26,7 +26,7 @@ def run(smoke_test=False, storage_path: str = None):
 
     config = (
         APPOConfig()
-        .environment("ale_py:ALE/Pong-v5", clip_rewards=True)
+        .environment("ALE/Pong-v5", clip_rewards=True)
         .framework(tune.grid_search(["tf", "torch"]))
         .rollouts(
             rollout_fragment_length=50,

release/ray_release/byod/requirements_byod_3.9.in

@@ -10,8 +10,8 @@ dask[complete]
 fastapi
 gcsfs==2023.5.0
 gsutil
-gymnasium
-gymnasium[atari]
+gym
+gym[atari]
 importlib-metadata
 jsonschema==4.17.3
 lightgbm