Use benchmark_environments test code (#16)

* Use benchmark_environments test code

* Force CI to regenerate cache

* Fix for Gym breaking change: openai/gym#1802

requirements.txt

@@ -1,12 +1,13 @@
-gym[mujoco]
+benchmark-environments @ git+https://github.com/HumanCompatibleAI/benchmark-environments.git
 imitation @ git+https://github.com/HumanCompatibleAI/imitation.git@e99844
+stable-baselines @ git+https://github.com/hill-a/stable-baselines.git
+gym[mujoco]
 matplotlib
 numpy
 pandas
 pymdptoolbox
 seaborn
 setuptools
 scipy
-stable-baselines @ git+https://github.com/hill-a/stable-baselines.git
 tensorflow>=1.13,<1.14
 xarray

setup.py

@@ -34,7 +34,7 @@ def load_requirements(fname):
     package_data={"evaluating_rewards": ["py.typed"]},
     install_requires=load_requirements("requirements.txt"),
     extras_require={"test": load_requirements("requirements-dev.txt")},
-    url="https://github.com/AdamGleave/evaluating_rewards",
+    url="https://github.com/HumanCompatibleAI/evaluating_rewards",
     license="Apache License, Version 2.0",
     classifiers=[
         # Trove classifiers

src/evaluating_rewards/envs/__init__.py

@@ -48,7 +48,7 @@ def register_similar(existing_name: str, new_name: str, **kwargs_delta):
         **kwargs_delta: Arguments to override the specification from existing_id.
     """
     existing_spec = gym.spec(existing_name)
-    fields = ["entry_point", "reward_threshold", "nondeterministic", "tags", "max_episode_steps"]
+    fields = ["entry_point", "reward_threshold", "nondeterministic", "max_episode_steps"]
     kwargs = {k: getattr(existing_spec, k) for k in fields}
     kwargs["kwargs"] = existing_spec._kwargs  # pylint:disable=protected-access
     kwargs.update(**kwargs_delta)

tests/common.py

@@ -17,7 +17,7 @@
 import copy
 from typing import Dict, Iterator, Tuple, TypeVar
 
-from imitation.testing import envs as test_envs
+from benchmark_environments.testing import envs as test_envs
 import pytest
 from stable_baselines.common import vec_env
 

tests/test_envs.py

@@ -18,8 +18,9 @@
 "evaluating_rewards/".
 """
 
+from benchmark_environments.testing import envs as bench_test
 import gym
-from imitation.testing import envs as test_envs
+from imitation.testing import envs as imitation_test
 import pytest
 
 from evaluating_rewards import envs  # noqa: F401 pylint:disable=unused-import
@@ -38,16 +39,22 @@ class TestEnvs:
     """Simple smoke tests for custom environments."""
 
     def test_seed(self, env, env_name):
-        test_envs.test_seed(env, env_name, DETERMINISTIC_ENVS)
+        bench_test.test_seed(env, env_name, DETERMINISTIC_ENVS)
 
-    def test_rollout(self, env):
-        test_envs.test_rollout(env)
+    def test_premature_step(self, env):
+        """Test that you must call reset() before calling step()."""
+        bench_test.test_premature_step(
+            env, skip_fn=pytest.skip, raises_fn=pytest.raises,
+        )
+
+    def test_rollout_schema(self, env):
+        bench_test.test_rollout_schema(env)
 
     def test_model_based(self, env):
         """Smoke test for each of the ModelBasedEnv methods with type checks."""
         if not hasattr(env, "state_space"):  # pragma: no cover
             pytest.skip("This test is only for subclasses of ModelBasedEnv.")
-        test_envs.test_model_based(env)
+        imitation_test.test_model_based(env)
 
 
 # pylint:enable=no-self-use