def test_subgoals():
"""Test that the right subgoals are set."""
environment = CombinationLock(3, 3, 6, 0.0, [1, 5, 4])
assert environment.num_states() == 9
assert environment.create_subgoals() == [Subgoal(2), Subgoal(5), Subgoal(8)]
def test_subgoals():
"""Test that the right subgoals are set."""
environment = CombinationLock(3, 3, 6, 0.0, [1, 5, 4])
assert environment.num_states() == 9
assert environment.create_subgoals() == [
Subgoal(2), Subgoal(5), Subgoal(8)
]
def main(argv):
"""Execute experiment."""
args = parse_args(argv)
random.seed(args.seed)
logging.basicConfig(level=log_level(args.log_level))
environment = (args.environment == 'gridworld' and Gridworld(args.gridworld_width, args.gridworld_height, args.failure_rate)) or \
(args.environment == 'gridworld_continuous' and GridworldContinuous(0.2, 0.01)) or \
(args.environment == 'combo_lock' and CombinationLock(args.gridworld_height, args.gridworld_width, 4, args.failure_rate))
policy = (args.agent_policy == 'random'
and RandomPolicy(environment.num_actions))
fa = ((args.environment == 'gridworld' or args.environment == 'combo_lock') and
TabularFA(environment.num_states(), environment.num_actions)) or \
(args.environment == 'gridworld_continuous' and RBF(2, 5, environment.num_actions, beta=args.beta))
agent = Agent(policy,
fa,
environment.num_actions,
args.alpha,
args.gamma,
args.eta,
args.zeta,
args.epsilon,
args.num_vi,
args.sim_samples,
args.sim_steps,
retain_theta=args.retain_theta,
subgoals=environment.create_subgoals())
agent.policy = RandomOptionPolicy(agent, args.random_options)
if args.agent_viz:
agent.create_visualization(
args.environment == 'gridworld'
or args.environment == 'combo_lock', environment)
# agent.exploit(np.asarray([1, 1]))
# agent.exploit(environment.num_states()-1)
# results_descriptor = ResultsDescriptor(args.results_interval, args.results_path, ['interval_id', 'steps'])
# experiment_descriptor = ExperimentDescriptor(args.plan_interval, args.num_steps)
# start(experiment_descriptor, agent, environment, results_descriptor)
e = Experiment2(agent, environment, args.plan_interval, args.num_steps,
args.viz_steps)
e.run()
def test_grid_position():
"""Are the grid positions correct?"""
environment = CombinationLock(3, 3, 6, [1, 5, 4], 0.0)
assert environment.grid_position_from_state(2) == GridPosition(2, 0)
assert environment.grid_position_from_state(3) == GridPosition(0, 1)
def test_taking_correct_actions():
"""Does the environment correctly change the state when told to take an action with and without stochasticity?"""
random.seed()
environment = CombinationLock(3, 3, 6, 0.0, [1, 5, 4])
assert environment.actions_from_state(0) == []
assert environment.actions_from_state(1) == [1]
assert environment.actions_from_state(2) == [1, 5]
assert environment.actions_from_state(3) == [1, 5, 4]
assert not environment.is_terminal([])
# Deterministic tests
state1 = environment.next_state(environment.initial_state(), 1)
assert state1 == 1
assert not environment.is_terminal(environment.actions_from_state(state1))
state2 = environment.next_state(state1, 5)
assert state2 == 2
assert not environment.is_terminal(environment.actions_from_state(state2))
state3 = environment.next_state(state2, 4)
assert state3 == 3
assert environment.is_terminal(environment.actions_from_state(state3) + [2])
state4 = environment.next_state(state3, 4)
assert state4 == 0
assert not environment.is_terminal(environment.actions_from_state(state4))
assert environment.is_terminal([2])
def test_grid_position():
"""Are the grid positions correct?"""
environment = CombinationLock(3, 3, 6, [1, 5, 4], 0.0)
assert environment.grid_position_from_state(2) == GridPosition(2, 0)
assert environment.grid_position_from_state(3) == GridPosition(0, 1)
def test_taking_correct_actions():
"""Does the environment correctly change the state when told to take an action with and without stochasticity?"""
random.seed()
environment = CombinationLock(3, 3, 6, 0.0, [1, 5, 4])
assert environment.actions_from_state(0) == []
assert environment.actions_from_state(1) == [1]
assert environment.actions_from_state(2) == [1, 5]
assert environment.actions_from_state(3) == [1, 5, 4]
assert not environment.is_terminal([])
# Deterministic tests
state1 = environment.next_state(environment.initial_state(), 1)
assert state1 == 1
assert not environment.is_terminal(environment.actions_from_state(state1))
state2 = environment.next_state(state1, 5)
assert state2 == 2
assert not environment.is_terminal(environment.actions_from_state(state2))
state3 = environment.next_state(state2, 4)
assert state3 == 3
assert environment.is_terminal(
environment.actions_from_state(state3) + [2])
state4 = environment.next_state(state3, 4)
assert state4 == 0
assert not environment.is_terminal(environment.actions_from_state(state4))
assert environment.is_terminal([2])
