def __init__(self, env: MazeWorldEpisodeLength):
left4 = Action(action=env.ACTIONS.LEFT)
left5 = Action(action=env.ACTIONS.LEFT)
up4 = Action(action=env.ACTIONS.UP)
up5 = Action(action=env.ACTIONS.UP)
choice1 = Choice()
choice2 = Choice()
left = Action(action=env.ACTIONS.LEFT)
right = Action(action=env.ACTIONS.RIGHT)
up = Action(action=env.ACTIONS.UP)
down = Action(action=env.ACTIONS.DOWN)
stop = Stop()
transitions = (
MachineRelation(left=Start(), right=choice1),
MachineRelation(left=choice1, right=left4),
MachineRelation(left=left4, right=left5, label=0),
MachineRelation(left=left5, right=choice2, label=0),
MachineRelation(left=left4, right=stop, label=1),
MachineRelation(left=left5, right=stop, label=1),
MachineRelation(left=choice1, right=up4),
MachineRelation(left=up4, right=up5, label=0),
MachineRelation(left=up5, right=choice2, label=0),
MachineRelation(left=up4, right=stop, label=1),
MachineRelation(left=up5, right=stop, label=1),
MachineRelation(left=choice2, right=left),
MachineRelation(left=choice2, right=right),
MachineRelation(left=choice2, right=up),
MachineRelation(left=choice2, right=down),
MachineRelation(
left=left,
right=stop,
label=1,
),
MachineRelation(left=right, right=stop, label=1),
MachineRelation(left=up, right=stop, label=1),
MachineRelation(left=down, right=stop, label=1),
MachineRelation(
left=left,
right=stop,
label=0,
),
MachineRelation(left=right, right=stop, label=0),
MachineRelation(left=up, right=stop, label=0),
MachineRelation(left=down, right=stop, label=0),
)
super().__init__(graph=MachineGraph(transitions=transitions))
def __init__(self, env: MazeWorldEpisodeLength):
stop = Stop()
up1 = Action(env.ACTIONS.UP)
choice = Choice()
transitions = [
MachineRelation(left=Start(), right=up1),
MachineRelation(left=up1, right=choice, label=0),
MachineRelation(left=choice, right=stop),
MachineRelation(left=choice, right=up1),
]
transitions = add_action_transitions(transitions)
super().__init__(graph=MachineGraph(transitions=transitions))
def main():
env = ArmEnvToggleTopOnly(size_x=5,
size_y=5,
cubes_cnt=4,
episode_max_length=600,
finish_reward=100,
action_minus_reward=-0.001,
tower_target_size=4)
vertexes = sorted([
Stop(),
Start(),
Action(env.ACTIONS.LEFT),
Action(env.ACTIONS.RIGHT),
Action(env.ACTIONS.UP),
Action(env.ACTIONS.DOWN),
# Action(env.ACTIONS.TOGGLE),
Choice(),
# Action(env.ACTIONS.LEFT),
# Action(env.ACTIONS.RIGHT),
# Action(env.ACTIONS.UP),
# Action(env.ACTIONS.DOWN),
# Action(env.ACTIONS.TOGGLE),
# Choice(),
])
# clearing directory
pathlib.Path('pics/').mkdir(parents=True, exist_ok=True)
shutil.rmtree('pics/')
pathlib.Path('pics/').mkdir(parents=True, exist_ok=True)
# brute force
for max_vertex_count in range(7):
vc = vertex_combination(vertex_types=vertexes,
max_vertex_count=max_vertex_count)
for index, vertex_types in enumerate(vc):
for graph_id in sorted(
get_graph_id_fast(
MachineStored(vertex_types=vertex_types,
binary_matrix_representation=412,
env=env))):
ms = MachineStored(vertex_types=vertex_types,
binary_matrix_representation=graph_id,
env=env)
if is_ham_ok(ms.get_machine_without_on_model()):
if check_for_one_component_graph(
ms.get_machine_without_on_model()):
if is_it_machine_runnable(
ms.get_machine_without_on_model()):
ms.draw("pics/" + str(max_vertex_count) + ":" +
str(index) + ":" + str(graph_id))
print("added")
def main():
env = ArmEnvToggleTopOnly(size_x=5, size_y=5, cubes_cnt=4, episode_max_length=600, finish_reward=100, action_minus_reward=-0.001, tower_target_size=4)
ms = MachineStored(vertex_types=sorted([
Start(),
Stop(),
Action(env.ACTIONS.LEFT),
Action(env.ACTIONS.LEFT),
Choice(),
Action(env.ACTIONS.RIGHT),
Action(env.ACTIONS.TOGGLE),
Action(env.ACTIONS.TOGGLE),
]), binary_matrix_representation=42, env=env)
ms.draw("a")
d = ms.to_dict()
ms = MachineStored.from_dict(d, env=env)
ms.draw("b")
def super_runner(call_me_maybe, env):
start = Start()
choice_one = Choice()
actions = [Action(action=_) for _ in env.get_actions_as_dict().values()]
stop = Stop()
call = Call(call_me_maybe)
transitions = [
MachineRelation(left=start, right=choice_one),
]
for action in actions:
transitions.append(MachineRelation(left=choice_one, right=action))
transitions.append(MachineRelation(left=action, right=stop, label=0))
transitions.append(MachineRelation(left=action, right=stop, label=1))
transitions.append(MachineRelation(left=choice_one, right=call))
transitions.append(MachineRelation(left=call, right=stop))
return AbstractMachine(graph=MachineGraph(transitions=transitions))
def from_dict(graph_dict, env):
vertex_types = []
for v in graph_dict["vertices"]:
if isinstance(v, (list, tuple)):
_, action_id = v
vertex_types.append(Action(action=action_id))
elif isinstance(v, str):
if v == "Choice":
vertex_types.append(Choice())
elif v == "Start":
vertex_types.append(Start())
elif v == "Stop":
vertex_types.append(Stop())
else:
raise TypeError
else:
raise TypeError
return MachineStored(vertex_types=vertex_types, binary_matrix_representation=graph_dict["binary_matrix_representation"], env=env)
def __init__(self, env):
start = Start()
choice_one = Choice()
actions = [
Action(action=_) for _ in env.get_actions_as_dict().values()
]
stop = Stop()
transitions = [
MachineRelation(left=start, right=choice_one),
]
for action in actions:
transitions.append(MachineRelation(left=choice_one, right=action))
transitions.append(
MachineRelation(left=action, right=stop, label=0))
transitions.append(
MachineRelation(left=action, right=stop, label=1))
super().__init__(graph=MachineGraph(transitions=transitions))
def main():
env = ArmEnvToggleTopOnly(size_x=5,
size_y=5,
cubes_cnt=4,
episode_max_length=600,
finish_reward=100,
action_minus_reward=-0.001,
tower_target_size=4)
vertex_types = sorted([
Stop(),
Start(),
Action(env.ACTIONS.LEFT),
Action(env.ACTIONS.RIGHT),
Action(env.ACTIONS.UP),
Action(env.ACTIONS.DOWN),
Call(None),
Choice(),
])
res = generate_list_of_vertexes(vertex_types=vertex_types,
vertex_of_each_type_max_count=3,
max_vertex_count=5)
for i in res:
print(vertex_list_to_str(i))
def main():
def get_on_model(self):
return self.get_arm_x(), self.is_cube_graped()
def get_all_on_model(self):
res = []
for height in range(0, self._size_x - 1):
for graped in [True, False]:
if height == self._size_x - 1 and graped:
continue
res.append((height, graped))
return res
def get_arm_x(self):
return self._arm_x
return self._size_x - self._arm_x
def is_cube_graped(self):
cube_dx, cube_dy = self.MOVE_ACTIONS[self.ACTIONS.DOWN]
cube_x, cube_y = self._arm_x + cube_dx, self._arm_y + cube_dy
return self._magnet_toggle and self.ok(
cube_x, cube_y) and self._grid[cube_x][cube_y] == 1
ArmEnvToggleTopOnly.get_arm_x = get_arm_x
ArmEnvToggleTopOnly.get_all_on_model = get_all_on_model
ArmEnvToggleTopOnly.is_cube_graped = is_cube_graped
ArmEnvToggleTopOnly.get_on_model = get_on_model
# env = ArmEnvToggleTopOnly(size_x=5, size_y=4, cubes_cnt=4, episode_max_length=500, finish_reward=100, action_minus_reward=-0.001, tower_target_size=4)
env = ArmEnvToggleTopOnly(size_x=4,
size_y=3,
cubes_cnt=3,
episode_max_length=200,
finish_reward=100,
action_minus_reward=-0.00001,
tower_target_size=3)
vertexes = sorted([
Stop(),
Start(),
Action(env.ACTIONS.LEFT),
Action(env.ACTIONS.RIGHT),
Action(env.ACTIONS.UP),
Action(env.ACTIONS.DOWN),
Action(env.ACTIONS.TOGGLE),
Choice(),
Action(env.ACTIONS.LEFT),
Action(env.ACTIONS.RIGHT),
Action(env.ACTIONS.UP),
Action(env.ACTIONS.DOWN),
Action(env.ACTIONS.TOGGLE),
# Choice(),
])
# part_one(env, vertexes)
# part_two(env)
# part_three(env)
# part_four(env)
env = ArmEnvToggleTopOnly(size_x=6,
size_y=4,
cubes_cnt=5,
episode_max_length=800,
finish_reward=100,
action_minus_reward=-0.00001,
tower_target_size=5)
# part_six(env)
# env = ArmEnvToggleTopOnly(size_x=5, size_y=4, cubes_cnt=4, episode_max_length=500, finish_reward=100, action_minus_reward=-0.001, tower_target_size=4)
part_seven(env)
def main():
class UpMachine4(AbstractMachine):
def __init__(self, env: ArmEnvToggleTopOnly):
d1 = Action(action=env.ACTIONS.UP)
d2 = Action(action=env.ACTIONS.UP)
d3 = Action(action=env.ACTIONS.UP)
d4 = Action(action=env.ACTIONS.UP)
stop = Stop()
transitions = (
MachineRelation(left=Start(), right=d1),
MachineRelation(left=d1, right=d2, label=0),
MachineRelation(left=d2, right=d3, label=0),
MachineRelation(left=d3, right=d4, label=0),
MachineRelation(left=d4, right=stop, label=0),
MachineRelation(left=d1, right=stop, label=1),
MachineRelation(left=d2, right=stop, label=1),
MachineRelation(left=d3, right=stop, label=1),
MachineRelation(left=d4, right=stop, label=1),
)
super().__init__(graph=MachineGraph(transitions=transitions))
class UpMachine3(AbstractMachine):
def __init__(self, env: ArmEnvToggleTopOnly):
d1 = Action(action=env.ACTIONS.UP)
d2 = Action(action=env.ACTIONS.UP)
d3 = Action(action=env.ACTIONS.UP)
# d4 = Action(action=env.ACTIONS.UP)
stop = Stop()
transitions = (
MachineRelation(left=Start(), right=d1),
MachineRelation(left=d1, right=d2, label=0),
MachineRelation(left=d2, right=d3, label=0),
MachineRelation(left=d3, right=stop, label=0),
# MachineRelation(left=d4, right=stop, label=0),
MachineRelation(left=d1, right=stop, label=1),
MachineRelation(left=d2, right=stop, label=1),
MachineRelation(left=d3, right=stop, label=1),
# MachineRelation(left=d4, right=stop, label=1),
)
super().__init__(graph=MachineGraph(transitions=transitions))
a = [
Choice(),
Action(ArmEnvToggleTopOnly.ACTIONS.RIGHT),
Action(ArmEnvToggleTopOnly.ACTIONS.LEFT),
Action(ArmEnvToggleTopOnly.ACTIONS.DOWN),
# Action(ArmEnvToggleTopOnly.ACTIONS.UP),
Call(machine_to_call=UpMachine4(environments[1])),
]
transitions = []
for i in a:
for j in a:
if randrange(2):
if isinstance(i, Action):
transitions.append(
MachineRelation(left=i, right=j, label=0))
else:
transitions.append(MachineRelation(left=i, right=j))
# len_ = len(goodhams)
# print(len_)
# len_4 = len_ // 4 + 1
# l1, r1 = 0, len_4
# l2, r2 = len_4, 2 * len_4
# l3, r3 = 2 * len_4, 3 * len_4
# l4, r4 = 3 * len_4, 4 * len_4
# print(l1, r1 )
# print(l2, r2 )
# print(l3, r3 )
# print(l4, r4 )
# exit(0)
# for brute_force in goodhams:
# for index, brute_force in enumerate(goodhams[l1: r1]):
# for index, brute_force in enumerate(goodhams[l2: r2]):
# for index, brute_force in enumerate(goodhams[l3: r3]):
brute_force = 1180698
# if bin(brute_force).count("1") > 12 or bin(brute_force).count("1") < 4:
# continue
# continue
go_continue = False
transitions = []
ss = set()
for ii in range(len(a)):
for jj in range(len(a)):
i = a[ii]
j = a[jj]
if (2**(ii * len(a) + jj)) & brute_force:
if isinstance(i, Action):
transitions.append(
MachineRelation(left=i, right=j, label=0))
else:
transitions.append(MachineRelation(left=i, right=j))
if ii in ss and isinstance(a[ii], (Action, Call)):
go_continue = True
break
ss.add(ii)
stop = Stop()
for ii in range(len(a)):
if ii not in ss:
i = a[ii]
if isinstance(i, Action):
transitions.append(MachineRelation(left=i, right=stop,
label=0))
else:
transitions.append(MachineRelation(left=i, right=stop))
for i in a:
if isinstance(i, Action):
transitions.append(MachineRelation(left=i, right=stop, label=1))
transitions.append(MachineRelation(left=Start(), right=a[0]))
machine = AbstractMachine(MachineGraph(transitions=transitions))
am = RootMachine(LoopInvokerMachine(machine))
env = environments[0]
draw_graph(
"{brute_force}".format(**locals()),
am.get_graph_to_draw(action_to_name_mapping=env.get_actions_as_dict()))
name = "02_auto"
def run(global_env):
full_name = name
params = HAMParamsCommon(environments[0])
ham_runner(ham=am,
num_episodes=global_episodes,
env=env,
params=params)
rewards = params.logs["ep_rewards"]
# with open(full_name + " cumulative_reward.txt", "w") as w:
# for out in get_cumulative_rewards(rewards=rewards):
# w.write(str(out) + '\n', )
with open(full_name + " reward.txt", "w") as w:
for out in rewards:
w.write(str(out) + '\n', )
def main():
# for global_env in EnvironmentsArticle().environments:
run(EnvironmentsArticle().environments[0])
if __name__ == '__main__':
main()
def __init__(self, env):
pull_up_start = Start()
pull_up_on = Action(action=env.get_actions_as_dict()["ON"])
pull_up_down_01 = Action(action=env.get_actions_as_dict()["DOWN"])
pull_up_down_02 = Action(action=env.get_actions_as_dict()["DOWN"])
pull_up_down_03 = Action(action=env.get_actions_as_dict()["DOWN"])
pull_up_down_04 = Action(action=env.get_actions_as_dict()["DOWN"])
pull_up_up_01 = Action(action=env.get_actions_as_dict()["UP"])
pull_up_up_02 = Action(action=env.get_actions_as_dict()["UP"])
pull_up_up_03 = Action(action=env.get_actions_as_dict()["UP"])
pull_up_up_04 = Action(action=env.get_actions_as_dict()["UP"])
pull_up_stop = Stop()
pull_up_transitions = (
MachineRelation(left=pull_up_start, right=pull_up_on),
MachineRelation(left=pull_up_on, right=pull_up_down_01, label=0),
MachineRelation(left=pull_up_down_01,
right=pull_up_down_02,
label=0),
MachineRelation(left=pull_up_down_02,
right=pull_up_down_03,
label=0),
MachineRelation(left=pull_up_down_03,
right=pull_up_down_04,
label=0),
MachineRelation(left=pull_up_down_04, right=pull_up_up_01,
label=0),
MachineRelation(left=pull_up_up_01, right=pull_up_up_02, label=0),
MachineRelation(left=pull_up_up_02, right=pull_up_up_03, label=0),
MachineRelation(left=pull_up_up_03, right=pull_up_up_04, label=0),
MachineRelation(left=pull_up_up_04, right=pull_up_stop, label=0),
MachineRelation(left=pull_up_on, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_01, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_02, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_03, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_04, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_01, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_02, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_03, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_04, right=pull_up_stop, label=1),
)
pull_up = AbstractMachine(
MachineGraph(transitions=pull_up_transitions))
start = Start()
choice_one = Choice()
left = Action(action=env.get_actions_as_dict()["LEFT"])
right = Action(action=env.get_actions_as_dict()["RIGHT"])
off = Action(action=env.get_actions_as_dict()["OFF"])
call = Call(machine_to_call=pull_up)
stop = Stop()
transitions = (
MachineRelation(left=start, right=choice_one),
MachineRelation(left=choice_one, right=left),
MachineRelation(left=choice_one, right=right),
MachineRelation(left=choice_one, right=off),
MachineRelation(left=choice_one, right=call),
MachineRelation(left=call, right=stop),
MachineRelation(left=left, right=stop, label=0),
MachineRelation(left=right, right=stop, label=0),
MachineRelation(left=off, right=stop, label=0),
MachineRelation(left=left, right=stop, label=1),
MachineRelation(left=right, right=stop, label=1),
MachineRelation(left=off, right=stop, label=1),
)
super().__init__(graph=MachineGraph(transitions=transitions))
MachineRelation(left=pull_up_up_03, right=pull_up_up_04, label=0),
MachineRelation(left=pull_up_up_04, right=pull_up_stop, label=0),
MachineRelation(left=pull_up_on, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_01, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_02, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_03, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_down_04, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_01, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_02, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_03, right=pull_up_stop, label=1),
MachineRelation(left=pull_up_up_04, right=pull_up_stop, label=1),
)
pull_up = AbstractMachine(MachineGraph(transitions=pull_up_transitions))
start = Start()
choice_one = Choice()
left = Action(action=env.get_actions_as_dict()["LEFT"])
right = Action(action=env.get_actions_as_dict()["RIGHT"])
off = Action(action=env.get_actions_as_dict()["OFF"])
call = Call(machine_to_call=pull_up)
stop = Stop()
transitions = (
MachineRelation(left=start, right=choice_one),
MachineRelation(left=choice_one, right=left),
MachineRelation(left=choice_one, right=right),
MachineRelation(left=choice_one, right=off),
MachineRelation(left=choice_one, right=call),
MachineRelation(left=call, right=stop),
def main():
class UpMachine4(AbstractMachine):
def __init__(self, env: ArmEnvToggleTopOnly):
d1 = Action(action=env.ACTIONS.UP)
d2 = Action(action=env.ACTIONS.UP)
d3 = Action(action=env.ACTIONS.UP)
d4 = Action(action=env.ACTIONS.UP)
stop = Stop()
transitions = (
MachineRelation(left=Start(), right=d1),
MachineRelation(left=d1, right=d2, label=0),
MachineRelation(left=d2, right=d3, label=0),
MachineRelation(left=d3, right=d4, label=0),
MachineRelation(left=d4, right=stop, label=0),
MachineRelation(left=d1, right=stop, label=1),
MachineRelation(left=d2, right=stop, label=1),
MachineRelation(left=d3, right=stop, label=1),
MachineRelation(left=d4, right=stop, label=1),
)
super().__init__(graph=MachineGraph(transitions=transitions))
class UpMachine3(AbstractMachine):
def __init__(self, env: ArmEnvToggleTopOnly):
d1 = Action(action=env.ACTIONS.UP)
d2 = Action(action=env.ACTIONS.UP)
d3 = Action(action=env.ACTIONS.UP)
# d4 = Action(action=env.ACTIONS.UP)
stop = Stop()
transitions = (
MachineRelation(left=Start(), right=d1),
MachineRelation(left=d1, right=d2, label=0),
MachineRelation(left=d2, right=d3, label=0),
MachineRelation(left=d3, right=stop, label=0),
# MachineRelation(left=d4, right=stop, label=0),
MachineRelation(left=d1, right=stop, label=1),
MachineRelation(left=d2, right=stop, label=1),
MachineRelation(left=d3, right=stop, label=1),
# MachineRelation(left=d4, right=stop, label=1),
)
super().__init__(graph=MachineGraph(transitions=transitions))
a = [
Choice(),
Action(ArmEnvToggleTopOnly.ACTIONS.RIGHT),
Action(ArmEnvToggleTopOnly.ACTIONS.LEFT),
Action(ArmEnvToggleTopOnly.ACTIONS.DOWN),
# Action(ArmEnvToggleTopOnly.ACTIONS.UP),
Call(machine_to_call=UpMachine4(environments[1])),
]
transitions = []
for i in a:
for j in a:
if randrange(2):
if isinstance(i, Action):
transitions.append(
MachineRelation(left=i, right=j, label=0))
else:
transitions.append(MachineRelation(left=i, right=j))
len_ = len(goodhams)
print(len_)
len_4 = len_ // 4 + 1
l1, r1 = 0, len_4
l2, r2 = len_4, 2 * len_4
l3, r3 = 2 * len_4, 3 * len_4
l4, r4 = 3 * len_4, 4 * len_4
# print(l1, r1 )
# print(l2, r2 )
# print(l3, r3 )
# print(l4, r4 )
# exit(0)
# for brute_force in goodhams:
# for index, brute_force in enumerate(goodhams[l1: r1]):
# for index, brute_force in enumerate(goodhams[l2: r2]):
# for index, brute_force in enumerate(goodhams[l3: r3]):
for index, brute_force in enumerate(goodhams[l4:r4]):
if index >= len_:
break
if index % (len_ // 100) == 0:
print(index // (len_ // 100), "%")
if bin(brute_force).count("1") > 10:
continue
if bin(brute_force).count("1") < 4:
continue
# if bin(brute_force).count("1") > 12 or bin(brute_force).count("1") < 4:
# continue
# continue
go_continue = False
transitions = []
ss = set()
for ii in range(len(a)):
for jj in range(len(a)):
i = a[ii]
j = a[jj]
if (2**(ii * len(a) + jj)) & brute_force:
if isinstance(i, Action):
transitions.append(
MachineRelation(left=i, right=j, label=0))
else:
transitions.append(MachineRelation(left=i, right=j))
if ii in ss and isinstance(a[ii], (Action, Call)):
go_continue = True
break
ss.add(ii)
if go_continue:
# print('continue')
continue
stop = Stop()
for ii in range(len(a)):
if ii not in ss:
i = a[ii]
if isinstance(i, Action):
transitions.append(
MachineRelation(left=i, right=stop, label=0))
else:
transitions.append(MachineRelation(left=i, right=stop))
for i in a:
if isinstance(i, Action):
transitions.append(MachineRelation(left=i, right=stop,
label=1))
for index_, II in enumerate(a):
transitions.append(MachineRelation(left=Start(), right=II))
go(transitions=transitions, brute_force=brute_force, index_=index_)
transitions.pop()