[RLlib-contrib] ES (evolutionary strategies). (#36625)

.buildkite/pipeline.ml.yml

@@ -456,7 +456,7 @@
     - conda deactivate
     - conda create -n rllib_contrib python=3.8 -y
     - conda activate rllib_contrib
-    - (cd rllib_contrib/a2c && pip install -r requirements.txt && pip install -e ".[development"])
+    - (cd rllib_contrib/a2c && pip install -r requirements.txt && pip install -e ".[development]")
     - ./ci/env/env_info.sh
     - pytest rllib_contrib/a2c/tests/
     - python rllib_contrib/a2c/examples/a2c_cartpole_v1.py --run-as-test
@@ -468,7 +468,7 @@
     - conda deactivate
     - conda create -n rllib_contrib python=3.8 -y
     - conda activate rllib_contrib
-    - (cd rllib_contrib/a3c && pip install -r requirements.txt && pip install -e ".[development"])
+    - (cd rllib_contrib/a3c && pip install -r requirements.txt && pip install -e ".[development]")
     - ./ci/env/env_info.sh
     - pytest rllib_contrib/a3c/tests/test_a3c.py
 
@@ -479,7 +479,7 @@
     - conda deactivate
     - conda create -n rllib_contrib python=3.8 -y
     - conda activate rllib_contrib
-    - (cd rllib_contrib/alpha_star && pip install -r requirements.txt && pip install -e ".[development"])
+    - (cd rllib_contrib/alpha_star && pip install -r requirements.txt && pip install -e ".[development]")
     - ./ci/env/env_info.sh
     - pytest rllib_contrib/alpha_star/tests/
     - python rllib_contrib/alpha_star/examples/multi-agent-cartpole-alpha-star.py --run-as-test
@@ -544,6 +544,18 @@
     - pytest rllib_contrib/ddpg/tests/
     - python rllib_contrib/ddpg/examples/ddpg_pendulum_v1.py --run-as-test
 
+- label: ":exploding_death_star: RLlib Contrib: ES Tests"
+  conditions: ["NO_WHEELS_REQUIRED", "RAY_CI_RLLIB_CONTRIB_AFFECTED"]
+  commands:
+    - cleanup() { if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then ./ci/build/upload_build_info.sh; fi }; trap cleanup EXIT
+    - conda deactivate
+    - conda create -n rllib_contrib python=3.8 -y
+    - conda activate rllib_contrib
+    - (cd rllib_contrib/es && pip install -r requirements.txt && pip install -e ".[development]")
+    - ./ci/env/env_info.sh
+    - pytest rllib_contrib/es/tests/
+    - python rllib_contrib/es/examples/es_cartpole_v1.py --run-as-test
+
 - label: ":exploding_death_star: RLlib Contrib: Leela Chess Zero Tests"
   conditions: ["NO_WHEELS_REQUIRED", "RAY_CI_RLLIB_CONTRIB_AFFECTED"]
   commands:
@@ -632,10 +644,11 @@
     - python rllib_contrib/r2d2/examples/r2d2_stateless_cartpole.py --run-as-test
 
 - label: ":exploding_death_star: RLlib Contrib: SimpleQ Tests"
-    - (cd rllib_contrib/simple_q && pip install -r requirements.txt && pip install -e ".[development]")
+    - cleanup() { if [ "${BUILDKITE_PULL_REQUEST}" = "false" ]; then ./ci/build/upload_build_info.sh; fi }; trap cleanup EXIT
     - conda deactivate
     - conda create -n rllib_contrib python=3.8 -y
     - conda activate rllib_contrib
+    - (cd rllib_contrib/simple_q && pip install -r requirements.txt && pip install -e ".[development]")
     - ./ci/env/env_info.sh
     - pytest rllib_contrib/simple_q/tests/
     - python rllib_contrib/simple_q/examples/simple_q_cartpole_v1.py --run-as-test

rllib_contrib/es/README.md

@@ -0,0 +1,17 @@
+# ES (Evolution Strategies)
+
+[ES](https://arxiv.org/abs/1703.03864) is a class of black box optimization algorithms, as an alternative to popular MDP-based RL techniques such as Q-learning and Policy Gradients. It is invariant to action frequency and delayed rewards, tolerant of extremely long horizons, and does not need temporal discounting or value function approximation.
+
+
+## Installation
+
+```
+conda create -n rllib-es python=3.10
+conda activate rllib-es
+pip install -r requirements.txt
+pip install -e '.[development]'
+```
+
+## Usage
+
+[ES Example]()

rllib_contrib/es/examples/es_cartpole_v1.py

@@ -0,0 +1,50 @@
+import argparse
+
+from rllib_es.es import ES, ESConfig
+
+import ray
+from ray import air, tune
+from ray.rllib.utils.test_utils import check_learning_achieved
+
+
+def get_cli_args():
+    """Create CLI parser and return parsed arguments"""
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--run-as-test", action="store_true", default=False)
+    args = parser.parse_args()
+    print(f"Running with following CLI args: {args}")
+    return args
+
+
+if __name__ == "__main__":
+    args = get_cli_args()
+
+    ray.init()
+
+    config = (
+        ESConfig()
+        .rollouts(num_rollout_workers=2)
+        .framework("torch")
+        .environment("CartPole-v1")
+        .training(noise_size=25000000, episodes_per_batch=50)
+    )
+
+    stop_reward = 100
+
+    tuner = tune.Tuner(
+        ES,
+        param_space=config.to_dict(),
+        run_config=air.RunConfig(
+            stop={
+                "sampler_results/episode_reward_mean": stop_reward,
+                "timesteps_total": 500000,
+            },
+            failure_config=air.FailureConfig(fail_fast="raise"),
+        ),
+    )
+    results = tuner.fit()
+
+    if args.run_as_test:
+        check_learning_achieved(
+            results, stop_reward, metric="sampler_results/episode_reward_mean"
+        )

rllib_contrib/es/pyproject.toml

@@ -0,0 +1,18 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+
+[tool.setuptools.packages.find]
+where = ["src"]
+
+[project]
+name = "rllib-es"
+authors = [{name = "Anyscale Inc."}]
+version = "0.1.0"
+description = ""
+readme = "README.md"
+requires-python = ">=3.7, <3.11"
+dependencies = ["gymnasium", "ray[rllib]==2.5.0"]
+
+[project.optional-dependencies]
+development = ["pytest>=7.2.2", "pre-commit==2.21.0", "tensorflow==2.11.0", "torch==1.12.0"]

rllib_contrib/es/requirements.txt

@@ -0,0 +1,2 @@
+tensorflow==2.11.0
+torch==1.12.0

rllib_contrib/es/src/rllib_es/es/__init__.py

@@ -0,0 +1,9 @@
+from rllib_es.es.es import ES, ESConfig
+from rllib_es.es.es_tf_policy import ESTFPolicy
+from rllib_es.es.es_torch_policy import ESTorchPolicy
+
+from ray.tune.registry import register_trainable
+
+__all__ = ["ES", "ESConfig", "ESTFPolicy", "ESTorchPolicy"]
+
+register_trainable("rllib-contrib-es", ES)