from pathlib import Path
from dacbench.logger import Logger
from dacbench.wrappers import PerformanceTrackingWrapper, ObservationWrapper
from examples.example_utils import make_chainer_a3c
from dacbench.benchmarks import CMAESBenchmark
# Make logger object
logger = Logger(experiment_name="CMAESBenchmark",
output_path=Path("../plotting/data"))
# Make CMA-ES environment
# We use the configuration from the "Learning to Optimize Step-size Adaption in CMA-ES" Paper by Shala et al.
bench = CMAESBenchmark()
env = bench.get_benchmark()
logger.set_env(env)
# Wrap to track performance
performance_logger = logger.add_module(PerformanceTrackingWrapper)
env = PerformanceTrackingWrapper(env=env, logger=performance_logger)
# Also wrap to make the dictionary observations into an easy to work with list
env = ObservationWrapper(env)
# Make chainer agent
obs_size = env.observation_space.low.size
action_size = env.action_space.low.size
agent = make_chainer_a3c(obs_size, action_size)
# Training
num_episodes = 3
for i in range(num_episodes):
def test_benchmark_env(self):
bench = CMAESBenchmark()
env = bench.get_benchmark()
self.assertTrue(issubclass(type(env), CMAESEnv))
def get_test_env(self) -> AbstractEnv:
bench = CMAESBenchmark()
env = bench.get_benchmark(seed=42)
return env