wehub-resource-sync
144 lines
4.6 KiB
Python
144 lines
4.6 KiB
Python
import unittest
|
|
|
|
import ray
|
|
from ray.rllib.algorithms import PPOConfig
|
|
from ray.rllib.callbacks.callbacks import RLlibCallback
|
|
|
|
|
|
class TestMultiCallback(unittest.TestCase):
|
|
"""A tests suite to test the `MultiCallback`."""
|
|
|
|
@classmethod
|
|
def setUp(cls) -> None:
|
|
ray.init()
|
|
|
|
@classmethod
|
|
def tearDown(cls) -> None:
|
|
ray.shutdown()
|
|
|
|
def test_multicallback_with_custom_callback_function(self):
|
|
"""Tests if callbacks in `MultiCallback` get executed.
|
|
|
|
This also tests, if multiple callbacks from different sources, i.e.
|
|
`callback_class` and `on_episode_step` run correctly.
|
|
"""
|
|
# Define two standard `RLlibCallback`.
|
|
class TestRLlibCallback1(RLlibCallback):
|
|
def on_episode_step(
|
|
self,
|
|
*,
|
|
episode,
|
|
env_runner=None,
|
|
metrics_logger=None,
|
|
env=None,
|
|
env_index,
|
|
rl_module=None,
|
|
worker=None,
|
|
base_env=None,
|
|
policies=None,
|
|
**kwargs
|
|
):
|
|
|
|
metrics_logger.log_value("callback_1", 1, reduce="mean")
|
|
|
|
class TestRLlibCallback2(RLlibCallback):
|
|
def on_episode_step(
|
|
self,
|
|
*,
|
|
episode,
|
|
env_runner=None,
|
|
metrics_logger=None,
|
|
env=None,
|
|
env_index,
|
|
rl_module=None,
|
|
worker=None,
|
|
base_env=None,
|
|
policies=None,
|
|
**kwargs
|
|
):
|
|
|
|
metrics_logger.log_value("callback_2", 2, reduce="mean")
|
|
|
|
# Define a custom callback function.
|
|
def custom_on_episode_step_callback(
|
|
episode,
|
|
env_runner=None,
|
|
metrics_logger=None,
|
|
env=None,
|
|
env_index=None,
|
|
rl_module=None,
|
|
worker=None,
|
|
base_env=None,
|
|
policies=None,
|
|
**kwargs
|
|
):
|
|
|
|
metrics_logger.log_value("custom_callback", 3, reduce="mean")
|
|
|
|
# Configure the algorithm.
|
|
config = (
|
|
PPOConfig()
|
|
.environment("CartPole-v1")
|
|
.api_stack(
|
|
enable_env_runner_and_connector_v2=True,
|
|
enable_rl_module_and_learner=True,
|
|
)
|
|
# Use the callbacks and callback function.
|
|
.callbacks(
|
|
callbacks_class=[TestRLlibCallback1, TestRLlibCallback2],
|
|
on_episode_step=custom_on_episode_step_callback,
|
|
)
|
|
)
|
|
|
|
# Build the algorithm. At this stage, callbacks get already validated.
|
|
algo = config.build()
|
|
|
|
# Run 10 training iteration and check, if the metrics defined in the
|
|
# callbacks made it into the results. Furthermore, check, if the values are correct.
|
|
for _ in range(10):
|
|
results = algo.train()
|
|
self.assertIn("callback_1", results["env_runners"])
|
|
self.assertIn("callback_2", results["env_runners"])
|
|
self.assertIn("custom_callback", results["env_runners"])
|
|
self.assertAlmostEqual(results["env_runners"]["callback_1"], 1)
|
|
self.assertAlmostEqual(results["env_runners"]["callback_2"], 2)
|
|
self.assertAlmostEqual(results["env_runners"]["custom_callback"], 3)
|
|
|
|
algo.stop()
|
|
|
|
def test_multicallback_validation_error(self):
|
|
"""Check, if the validation safeguard catches wrong `MultiCallback`s."""
|
|
with self.assertRaises(ValueError):
|
|
(
|
|
PPOConfig()
|
|
.environment("CartPole-v1")
|
|
.api_stack(
|
|
enable_env_runner_and_connector_v2=True,
|
|
enable_rl_module_and_learner=True,
|
|
)
|
|
# This is wrong b/c it needs callables.
|
|
.callbacks(callbacks_class=["TestRLlibCallback1", "TestRLlibCallback2"])
|
|
)
|
|
|
|
def test_single_callback_validation_error(self):
|
|
"""Tests if the validation safeguard catches wrong `RLlibCallback`s."""
|
|
with self.assertRaises(ValueError):
|
|
(
|
|
PPOConfig()
|
|
.environment("CartPole-v1")
|
|
.api_stack(
|
|
enable_env_runner_and_connector_v2=True,
|
|
enable_rl_module_and_learner=True,
|
|
)
|
|
# This is wrong b/c it needs callables.
|
|
.callbacks(callbacks_class="TestRLlibCallback")
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
import sys
|
|
|
|
import pytest
|
|
|
|
sys.exit(pytest.main(["-v", __file__]))
|