def get_machine(self):
transitions = []
for left_ind in range(len(self.vertex_types)):
for right_ind in range(len(self.vertex_types)):
left = self.vertex_types[left_ind]
right = self.vertex_types[right_ind]
if (2 ** (left_ind * len(self.vertex_types) + right_ind)) & self.binary_matrix_representation:
if isinstance(left, Action):
transitions.append(MachineRelation(left=left, right=right, label=0))
else:
transitions.append(MachineRelation(left=left, right=right))
start, stop = None, None
for vertex in self.vertex_types:
if isinstance(vertex, Start):
start = vertex
elif isinstance(vertex, Stop):
stop = vertex
assert start is not None
assert stop is not None
for vertex in [_ for _ in self.vertex_types if isinstance(_, Action)]:
transitions.append(MachineRelation(left=vertex, right=stop, label=1))
return AbstractMachine(MachineGraph(transitions=transitions, vertices=self.vertex_types))
def go(transitions, brute_force, index_):
machine = AbstractMachine(MachineGraph(transitions=transitions))
am = RootMachine(LoopInvokerMachine(machine))
# if randrange(1000) == 0:
# draw_graph("{brute_force}".format(**locals()), am.get_graph_to_draw(action_to_name_mapping=env.get_actions_as_dict()))
# exit(0)
if is_it_machine_runnable(machine):
sum_rew = 0
try:
params = HAMParamsCommon(environments[0])
ham_runner(ham=am,
num_episodes=2,
env=environments[0],
params=params)
sum_rew = sum(params.logs["ep_rewards"])
except ChildProcessError:
# print(brute_force)
pass
# if randrange(1500) == 0:
# draw_graph("bf{brute_force}".format(**locals()), am.get_graph_to_draw())
if sum_rew > 0:
# TODO
# with open("out.txt", "a") as f:
# f.write(str(brute_force) + "\n")
# return
# print("\n\n EPISODE REWARD: ", sum_rew)
# draw_graph("{sum_rew}__{brute_force}".format(**locals()), am.get_graph_to_draw(action_to_name_mapping=env.get_actions_as_dict()))
rew = None
print("\n\n\n")
for e in environments:
params = HAMParamsCommon(e)
ham_runner(ham=am, num_episodes=600, env=e, params=params)
if rew is None:
rew = 0
rew += sum(params.logs["ep_rewards"])
print("to_add:", sum(params.logs["ep_rewards"]))
# except ChildProcessError:
# draw_graph("{rew}__{brute_force}".format(**locals()), am.get_graph_to_draw(action_to_name_mapping=env.get_actions_as_dict()))
# exit(0)
# pass
if rew is not None:
draw_graph(
"{rew}__{brute_force}_{index_}".format(**locals()),
am.get_graph_to_draw(
action_to_name_mapping=env.get_actions_as_dict()))
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 __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__(machine_to_invoke=AbstractMachine(
MachineGraph(transitions=transitions)))
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_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),