def test_score_agent(config):
"""Smoke test for score agent to check it runs with some different configs."""
config = dict(config)
if "episodes" not in config:
config["episodes"] = 1 # speed up tests
config[
"render"] = False # faster without, test_experiment already tests with render
run = score_ex.run(config_updates=config)
assert run.status == "COMPLETED"
outcomes = [run.result[k] for k in ["ties", "win0", "win1"]]
assert sum(outcomes) == run.config["episodes"]
if config.get("record_traj", False):
try:
for i in range(2):
traj_file_path = os.path.join(
config["record_traj_params"]["save_dir"], f"agent_{i}.npz")
traj_data = np.load(traj_file_path)
assert set(traj_data.keys()).issuperset(
["observations", "actions", "rewards"])
for k, ep_data in traj_data.items():
assert len(ep_data) == config[
"episodes"], f"unexpected array length at '{k}'"
os.remove(traj_file_path)
finally:
os.rmdir(config["record_traj_params"]["save_dir"])
def score_worker(base_config, tune_config, reporter):
"""Run a aprl.score experiment with specified config, logging to reporter.
:param base_config: (dict) default config
:param tune_config: (dict) overrides values in base_config
:param reporter: (ray.tune.StatusReporter) Ray Tune internal logger."""
common_worker.fix_sacred_capture()
# score_ex is not pickleable, so we cannot close on it.
# Instead, import inside the function.
from aprl.score_agent import score_ex
config = dict(base_config)
tune_config = common_worker.flatten_config(tune_config)
common_worker.update(config, tune_config)
# We're breaking the Sacred interface by running an experiment from within another experiment.
# This is the best thing we can do, since we need to run the experiment with varying configs.
# Just be careful: this could easily break things.
observer = observers.FileStorageObserver(
osp.join("data", "sacred", "score"))
score_ex.observers.append(observer)
run = score_ex.run(config_updates=config)
index_keys = config.get("index_keys", [])
idx = {
k: v
for k, v in config.items()
if k.startswith("agent") or k == "env_name" or k in index_keys
}
reporter(done=True, score=run.result, idx=idx)
def test_score_agent_video():
# Confirm that experiment runs properly saving videos to a temp dir
none_dir_run = score_ex.run(
config_updates=SCORE_AGENT_VIDEO_CONFIGS["none_dir"])
assert none_dir_run.status == "COMPLETED"
try:
# Confirm that the first time you try to save videos to a specified dir, it works properly
specified_dir_run = score_ex.run(
config_updates=SCORE_AGENT_VIDEO_CONFIGS["specified_dir"])
assert specified_dir_run.status == "COMPLETED"
# Confirm that the second time you try to save videos to the same specified dir, it fails
with pytest.raises(AssertionError):
_ = score_ex.run(
config_updates=SCORE_AGENT_VIDEO_CONFIGS["specified_dir"])
finally:
shutil.rmtree(SCORE_AGENT_VIDEO_CONFIGS["specified_dir"]
["video_params"]["save_dir"])
