def run_simulation(sim_vals) -> float:
"""
Run caterpillar with a policy given in argument.
This function is for multiprocessing.
sim_vals: (
steps: int,
actor_module_name: str,
actor_params: [np.array(params_0), np.array(params_1), ...],
disable_list: list
)
"""
steps, actor_module_name, actor_params, disable_list = sim_vals
assert isinstance(steps, int)
assert isinstance(actor_params, Iterable)
assert isinstance(disable_list, Iterable)
# Init actor
actor_module = import_module(actor_module_name)
actor = getattr(actor_module, config.COMMON_ACTOR_NAME)()
actor.set_params(actor_params)
# Init caterpillar
caterpillar = Caterpillar(config.somites, mode="default")
# Run steps
(accumulated_tension,) = exec_steps(steps, actor, caterpillar, disable_list=disable_list)
reward = caterpillar.moved_distance() - accumulated_tension / config.params["tension_divisor"]
return reward
def exec_steps(steps: int, actor: base_actor.BaseActor, caterpillar: Caterpillar,
disable_list: list, episode=None) -> float:
"""
Run caterpillar for designated steps.
return run information which is (accumulated tensions,)
"""
accumulated_tension = 0
for step in range(steps):
(_, _, tensions), _ = caterpillar_runner.observe_and_act(actor, caterpillar, disable_list, episode=episode)
accumulated_tension += np.sum(np.power(tensions, 2))
caterpillar.step(step, int(1 / config.params["time_dalte"] / 10))
return (accumulated_tension,)
def run_caterpillar(actor,
save_dir: str,
steps: int,
disable_list=None,
broken_value=0):
if disable_list is None:
disable_list = []
caterpillar = Caterpillar(config.somites, mode="default")
# Record during sampling
sim_distance_file = DataCSVSaver(os.path.join(save_dir, "distance.txt"),
("step", "distance"))
sim_phase_diffs_file = DataCSVSaver(
os.path.join(save_dir, "phase_diffs.txt"),
["step"] + ["phi_{}".format(i) for i in range(config.oscillators)])
sim_phases_file = DataCSVSaver(
os.path.join(save_dir, "phases.txt"),
["step"] + ["phi_{}".format(i) for i in range(config.oscillators)])
sim_actions_file = DataCSVSaver(
os.path.join(save_dir, "actions.txt"),
["step"] + ["action_{}".format(i) for i in range(config.oscillators)])
sim_frictions_file = DataCSVSaver(
os.path.join(save_dir, "frictions.txt"),
["step"] + ["friction_{}".format(i) for i in range(config.somites)])
sim_tension_file = DataCSVSaver(
os.path.join(save_dir, "tensions.txt"),
["step"] + ["tension_{}".format(i) for i in range(config.oscillators)])
for step in range(steps):
obv, action = observe_and_act(actor,
caterpillar,
disable_list=disable_list,
broken_value=broken_value)
_, frictions, tensions = obv
caterpillar.step(step, int(1 / config.params["time_dalte"] / 10))
# Save data
sim_distance_file.append_data(step, caterpillar.moved_distance())
sim_phase_diffs_file.append_data(step, *caterpillar.phases_from_base())
sim_phases_file.append_data(step, *caterpillar.phis)
sim_actions_file.append_data(step, *action)
sim_frictions_file.append_data(step, *frictions)
sim_tension_file.append_data(step, *tensions)
caterpillar.save_simulation("{}/render.sim".format(save_dir))
return caterpillar.moved_distance()
def train_caterpillar(save_dir: utils.SaveDir, actor_module_name: str):
# Dump train parameters
config.print_config()
actor_module = import_module(actor_module_name)
actor_class = getattr(actor_module, config.COMMON_ACTOR_NAME)
rlm = PEPGActorManager(actor_class)
caterpillar = Caterpillar(config.somites, mode="default")
config.dump_config(save_dir.log_dir())
distance_log = DataCSVSaver(os.path.join(save_dir.log_dir(), "distance.txt"), ("episode", "distance"))
reward_log = DataCSVSaver(os.path.join(save_dir.log_dir(), "reward.txt"), ("episode", "reward"))
sigma_log = DataCSVSaver(os.path.join(save_dir.log_dir(), "sigma.txt"), ("episode", "average sigma"))
episode = 0
while episode < config.params["episodes"]:
params_sets = rlm.sample_params()
print("\nEpisode: {}".format(episode))
print("---------------------------------------------------------------------")
rewards = []
try:
with Pool(processes=config.exec_params["worker_processes"], initializer=mute) as pool:
rewards = pool.map(run_simulation, [(config.params["steps"], actor_module_name, p_set, []) for p_set in params_sets])
rlm.update_params(np.array(rewards))
episode += 1
# Try parameters after this episode --------------------------------------
rlm.set_params(rlm.parameters)
caterpillar.reset()
(accumulated_tension,) = exec_steps(config.params["steps"], rlm.get_actor(), caterpillar, [], episode=episode - 1)
reward = caterpillar.moved_distance() - accumulated_tension / config.params["tension_divisor"]
# Save parameter performance
distance_log.append_data(episode, caterpillar.moved_distance())
sigma_log.append_data(episode, np.mean(rlm.sigmas))
reward_log.append_data(episode, reward)
announce = " --- Distance: {} Reward: {}".format(caterpillar.moved_distance(), reward)
print(announce)
except KeyboardInterrupt:
command = input("\nSample? Finish? : ")
if command in ["sample", "Sample"]:
rlm.set_params(rlm.parameters)
test_current_params(rlm.get_actor(), save_dir.log_dir(), episode)
continue
if command in ["finish", "Finish"]:
print("Ending training ...")
break
rlm.set_params(rlm.parameters)
rlm.save(save_file_path=os.path.join(save_dir.model_dir(), 'actor_model.pickle'))
return rlm.get_actor()
class InputCaterpillar(BoxLayout):
def __init__(self,**kwargs):
self.spacing=5
self.register_event_type('on_input')
super(InputCaterpillar, self).__init__(**kwargs)
self.orientation="vertical"
self.chain_widget=Caterpillar()
self.func=None
buttons=BoxLayout(orientation="horizontal")
for (k,v) in sorted(Caterpillar._colors.items()):
b=Button()
b.num=k
b.background_color=[c*3 for c in list(v)]+[1]
b.bind(on_press=self.color_press)
buttons.add_widget(b)
b=Button(text="<=")
b.bind(on_press=self.backspace)
buttons.add_widget(b)
b=Button(text=u"OK")
b.bind(on_press=self.enter)
buttons.add_widget(b)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
def on_input(self,chain):
pass
def set_chain(self,chain):
# if caterpillar is valid it looks to the left, otherwise to the right
eyes='right'
eyes='left' if chain and self.func and self.func(chain) else 'right'
self.chain_widget.set_chain(chain,eye_direction=eyes)
def backspace(self,instance):
chain=self.chain_widget.get_chain()[:-1]
self.set_chain(chain)
def color_press(self,instance):
chain=self.chain_widget.get_chain()
if len(chain)==7: return
chain=chain+[instance.num]
self.set_chain(chain)
def enter(self,instance):
if self.chain_widget.get_chain()==[]:return
self.dispatch('on_input',tuple(self.chain_widget.get_chain()))
self.chain_widget.set_chain([])
def test_current_params(actor: base_actor, log_dir: str, episode: int):
steps = input("How many steps for this sample?: ")
if steps == "":
utils.notice("default steps {}".format(config.params["default_sample_steps"]))
steps = config.params["default_sample_steps"]
# Record during sampling
sim_distance_file = DataCSVSaver(
"{}/train_result_ep{}_distance.txt".format(log_dir, episode),
("step", "distance")
)
sim_phase_diffs_file = DataCSVSaver(
"{}/train_result_ep{}_phase_diffs.txt".format(log_dir, episode),
["step"] + ["phi_{}".format(i) for i in range(config.oscillators)]
)
sim_actions_file = DataCSVSaver(
"{}/train_result_ep{}_actions.txt".format(log_dir, episode),
["step"] + ["action_{}".format(i) for i in range(config.oscillators)]
)
sim_frictions_file = DataCSVSaver(
"{}/train_result_ep{}_frictions.txt".format(log_dir, episode),
["step"] + ["friction_{}".format(i) for i in range(config.somites)]
)
caterpillar = Caterpillar(config.somites, mode="default")
for step in range(int(steps)):
try:
_, action = caterpillar_runner.observe_and_act(actor, caterpillar, [])
caterpillar.step(step, int(1 / config.params["time_dalte"] / 10))
except Exception:
continue
else:
# Save data
sim_distance_file.append_data(step, caterpillar.moved_distance())
sim_phase_diffs_file.append_data(step, *caterpillar.phases_from_base())
sim_actions_file.append_data(step, *action)
frictions = caterpillar.frictions()
sim_frictions_file.append_data(step, *frictions)
print("Moved distance:", caterpillar.moved_distance())
caterpillar.save_simulation("{}/train_result_ep{}.sim".format(log_dir, episode))
def draw(self, *args):
self.box.clear_widgets()
x, y = self.size
if x > y: cols = 2
else: cols = 1
self.box.cols = cols
for guess in self.caterpillars:
self.box.add_widget(
Caterpillar(chain=guess,
size_hint_y=None,
size=(x, x / 7 / cols)))
self.scroll_y = 1
def __init__(self, **kwargs):
self.register_event_type('on_input')
super(InputCaterpillar, self).__init__(**kwargs)
self.orientation = "vertical"
self.chain_widget = Caterpillar()
self.func = None
buttons = BoxLayout(orientation="horizontal")
for (k, v) in sorted(Caterpillar._colors.items()):
b = Button()
b.num = k
b.background_color = list(v) + [1]
b.bind(on_press=self.color_press)
buttons.add_widget(b)
b = Button(text="<=")
b.bind(on_press=self.backspace)
buttons.add_widget(b)
b = Button(text=u"OK")
b.bind(on_press=self.enter)
buttons.add_widget(b)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
def observe_and_act(actor: base_actor.BaseActor,
caterpillar: Caterpillar,
disable_list: list,
broken_value=0,
episode=None):
assert np.all(np.array(disable_list) < config.somites)
if disable_list is None:
disable_list = []
mask = np.ones(config.somites)
mask[disable_list] = 0
bias = np.zeros(config.somites)
bias[disable_list] = broken_value
frictions = caterpillar.frictions()
phis = caterpillar.phis
tensions = caterpillar.tensions()
state = np.concatenate([frictions * mask + bias, phis, tensions])
action = actor.get_action(state)
caterpillar.feedback_phis(action)
observation = (phis, frictions, tensions)
return observation, action
class InputCaterpillar(BoxLayout):
def __init__(self, **kwargs):
self.register_event_type('on_input')
super(InputCaterpillar, self).__init__(**kwargs)
self.orientation = "vertical"
self.chain_widget = Caterpillar()
self.func = None
buttons = BoxLayout(orientation="horizontal")
for (k, v) in sorted(Caterpillar._colors.items()):
b = Button()
b.num = k
b.background_color = list(v) + [1]
b.bind(on_press=self.color_press)
buttons.add_widget(b)
b = Button(text="<=")
b.bind(on_press=self.backspace)
buttons.add_widget(b)
b = Button(text=u"OK")
b.bind(on_press=self.enter)
buttons.add_widget(b)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
def on_input(self, chain):
pass
def set_chain(self, chain):
# if caterpillar is valid it looks to the left, otherwise to the right
eyes = 'right'
eyes = 'left' if chain and self.func and self.func(chain) else 'right'
self.chain_widget.set_chain(chain, eye_direction=eyes)
def backspace(self, instance):
chain = self.chain_widget.get_chain()[:-1]
self.set_chain(chain)
def color_press(self, instance):
chain = self.chain_widget.get_chain()
if len(chain) == 7: return
chain = chain + [instance.num]
self.set_chain(chain)
def enter(self, instance):
if self.chain_widget.get_chain() == []: return
self.dispatch('on_input', tuple(self.chain_widget.get_chain()))
self.chain_widget.set_chain([])
def __init__(self,**kwargs):
self.register_event_type('on_finish')
self.register_event_type('on_answer')
super(Exam, self).__init__(**kwargs)
self.orientation="vertical"
self.question_label=Label()
self.chain_widget=Caterpillar()
buttons=BoxLayout(orientation="horizontal")
b=Button(text="Valid")
b.background_color=(0,1,0,1)
b.val=True
b.bind(on_press=self.answer)
buttons.add_widget(b)
b=Button(text="Invalid")
b.background_color=(1,0,0,1)
b.val=False
b.bind(on_press=self.answer)
buttons.add_widget(b)
self.add_widget(self.question_label)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
def __init__(self,**kwargs):
self.register_event_type('on_input')
super(InputCaterpillar, self).__init__(**kwargs)
self.orientation="vertical"
self.chain_widget=Caterpillar()
self.func=None
buttons=BoxLayout(orientation="horizontal")
for (k,v) in sorted(Caterpillar._colors.items()):
b=Button()
b.num=k
b.background_color=list(v)+[1]
b.bind(on_press=self.color_press)
buttons.add_widget(b)
b=Button(text="<=")
b.bind(on_press=self.backspace)
buttons.add_widget(b)
b=Button(text=u"OK")
b.bind(on_press=self.enter)
buttons.add_widget(b)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
class Exam(BoxLayout):
def __init__(self,**kwargs):
self.register_event_type('on_finish')
self.register_event_type('on_answer')
super(Exam, self).__init__(**kwargs)
self.orientation="vertical"
self.question_label=Label()
self.chain_widget=Caterpillar()
buttons=BoxLayout(orientation="horizontal")
b=Button(text="Valid")
b.background_color=(0,1,0,1)
b.val=True
b.bind(on_press=self.answer)
buttons.add_widget(b)
b=Button(text="Invalid")
b.background_color=(1,0,0,1)
b.val=False
b.bind(on_press=self.answer)
buttons.add_widget(b)
self.add_widget(self.question_label)
self.add_widget(self.chain_widget)
self.add_widget(buttons)
def start_exam(self,valid,invalid,valid_history,invalid_history):
self.questions=self.make_questions(valid,invalid,valid_history,invalid_history)
self.question_num=0
self.ask_question()
def make_questions(self,valid,invalid,valid_history,invalid_history):
valid_num=random.randint(5,10)
invalid_num=15-valid_num
valids=[(x,True) for x in get_n(valid_num,valid,valid_history)]
invalids=[(x,False) for x in get_n(invalid_num,invalid,invalid_history)]
l=valids+invalids
random.shuffle(l)
return l
def ask_question(self):
if self.question_num==len(self.questions):
self.dispatch("on_finish",True)
else:
self.question_label.text="Question %s/%s"%(self.question_num+1,len(self.questions))
self.chain_widget.set_chain(self.questions[self.question_num][0])
def answer(self,instance):
self.dispatch("on_answer",self.questions[self.question_num][0])
if self.questions[self.question_num][1]==instance.val:
self.question_num+=1
self.ask_question()
else:
self.dispatch("on_finish",False)
def on_finish(self,*args):
pass
def on_answer(self,*args):
pass
def fight():
fight = True
#PROBABLY CHANGE
opponent = PaintbrushTown.challenger()
if opponent == 'Rocky':
enemy = 'bat'
elif opponent == 'Slander':
enemy = 'rat'
elif opponent == 'Kyle':
enemy = 'caterpillar'
print(opponent,'chooses', enemy,'\b!')
if(enemy == 'bat'):
combatant = Bat()
Bat.work()
elif(enemy == 'caterpillar'):
combatant = Caterpillar()
Caterpillar.work()
elif(enemy == 'rat'):
combatant = Rat()
Rat.work()
time.sleep(1)
while(True):
start,attack, = '0','0'
while start != 'k' and start != 'b' and start != 'm':
start = input('Which bokémon would you like to start with?\n[k = Kennen, b = Belldolpine, m = Meister Beast]\n').lower()
while(fight == True):
#PLAYER TURN
if start == 'k':
print('You have chosen Kennen!')
you = 'Kennen'
pc = Kennen()
elif start == 'b':
print('You have chosen Belldolpine!')
you = 'Belldolpine'
pc = Belldolpine()
elif start == 'm':
print('You have chosen Meister Beast!')
you = 'MeisterBeast'
pc = MeisterBeast()
while attack != 'b' and attack != 's':
attack = input('Would you like to use your basic attack [b] or your special attack [s]\n')
if attack == 'b':
print(you,'does', pc.basicDamage, 'damage to the', enemy)
combatant.takeDamage(pc.basicDamage)
attack = '0'
break
if attack == 's':
print(you, 'does', pc.specialDamage, 'damage to the', enemy)
combatant.takeDamage(pc.specialDamage)
attack = '0'
break
#ENEMY TURN
if(combatant.hp > 0):
time.sleep(1)
print('The', enemy, 'attacks!')
combatant.basicAttack()
print('The', enemy, 'does', combatant.basicAttack(), 'damage to you!')
pc.takeDamage(combatant.damage)
else:
time.sleep(1)
print('The', enemy, 'is knocked unconscious!\nYou have won!\n',opponent,'congradulates you and hands you a badge.')
time.sleep(1)
print('You should take the badge back to Professor Cottonwood so you can go find your parents.')
time.sleep(5)
pc.heal()
fight = False
break
