def setUp(self) -> None:
self.env = environment_impl.SingleStateEnvironment(
action_space_size=1, step_limit=10)
self.qfunc = qfunc_impl.MemoizationQFunction(
action_space_size=3,
discount_factor=0.9,
learning_rate=0.9)
self.runner = runner_impl.SimpleRunner()
def _RunEnv(gym_env):
env = screen_learning.ScreenGymEnvironment(gym_env)
qfunc = qfunc_impl.DQN_TargetNetwork(
model=screen_learning.CreateConvolutionModel(
action_space_size=env.GetActionSpaceSize()))
policy = policy_impl.GreedyPolicyWithRandomness(epsilon=1.0)
runner_impl.SimpleRunner().Run(
env=env, qfunc=qfunc, policy=policy, num_of_episodes=10)
def _RunEnv(gym_env):
env = environment_impl.GymEnvironment(gym_env)
qfunc = qfunc_impl.RandomValueQFunction(
action_space_size=env.GetActionSpaceSize())
env.Reset()
policy = policy_impl.GreedyPolicyWithRandomness(epsilon=1.0)
runner_impl.SimpleRunner().Run(
env=env, qfunc=qfunc, policy=policy, num_of_episodes=10)
def _RunEnv(gym_env):
env = environment_impl.GymEnvironment(gym_env)
env.SetGymEnvMaxEpisodeSteps(10)
qfunc = qfunc_impl.MemoizationQFunction(
action_space_size=env.GetActionSpaceSize())
env.Reset()
policy = policy_impl.GreedyPolicyWithRandomness(epsilon=1.0)
runner_impl.SimpleRunner().Run(env=env,
brain=qfunc,
policy=policy,
num_of_episodes=1)
def Demo(_):
env = screen_learning.ScreenGymEnvironment(gym.make('SpaceInvaders-v0'))
brain = a3c_impl.A3C(model=a3c_impl.CreateModel(
state_shape=env.GetStateShape(),
action_space_size=env.GetActionSpaceSize(),
hidden_layer_sizes=(12, ),
))
brain.Load('saved_models/a3c_invader.weights')
policy = policy_impl.GreedyPolicy()
env.StartRecording('a3c_invader.mp4')
runner = runner_impl.SimpleRunner()
runner.Run(env=env, brain=brain, policy=policy, num_of_episodes=10)
env.StopRecording()
def _RunEnv(gym_env):
env = environment_impl.GymEnvironment(gym_env)
env.SetGymEnvMaxEpisodeSteps(10)
qfunc = qfunc_impl.DQN(model=qfunc_impl.CreateModel(
state_shape=env.GetStateShape(),
action_space_size=env.GetActionSpaceSize(),
hidden_layer_sizes=(4, ),
))
env.Reset()
policy = policy_impl.GreedyPolicyWithRandomness(epsilon=1.0)
runner_impl.SimpleRunner().Run(env=env,
qfunc=qfunc,
policy=policy,
num_of_episodes=10)
def test_simpleRunner(self):
# Tests that it can run; quality if not important for this test.
runner_impl.SimpleRunner().Run(
env=self.env, qfunc=self.qfunc, policy=self.policy, num_of_episodes=1)