def reward(self, tcube):
# def reward(self):
if utils.perc_solved_cube(self.cube) == 1:
return 1000, True
return (utils.perc_solved_cube(self.cube) -
utils.perc_solved_cube(tcube)) * 100, False
transformation = [choice(list(action_map_small.keys())) for z in range(saa)]
my_formula = pc.Formula(transformation)
cube(my_formula)
run = True
print("********************************************************")
print("START : ")
print([cube])
print("********************************************************")
time.sleep(3)
while run:
if perc_solved_cube(cube) == 1:
run = False
print("Solved")
break
# Take the biggest Q value (= the best action)
flat_cube = np.array(flatten_1d_b(cube)).reshape([1, 324])
Qs = sess.run(myAgent.output, feed_dict = {myAgent.state_in: flat_cube})
# print("Ideal : "+str(my_formula.reverse()))
print("Q values:")
print(Qs)
# Take the biggest Q value (= the best action)
choice = np.argmax(Qs)
# lookup = ["R", "L","D","U","B","F","R'", "L'","D'","U'","B'","F'"] #We are not accounting for half turns
def solve(Cube):
file_path = "auto.h5"
model = get_model()
model.load_weights(file_path)
cube = Cube
cube.score = 0
list_sequences = [[cube]]
existing_cubes = set()
action_list = []
success = False
for j in range(50):
X = [flatten_1d_b(x[-1]) for x in list_sequences]
value, policy = model.predict(np.array(X), batch_size=1024)
new_list_sequences = []
for x, policy in zip(list_sequences, policy):
new_sequences = [x + [x[-1].copy()(action)] for action in action_map]
pred = np.argsort(policy)
take_action = list(action_map.keys())[pred[-1]]
# append action
action_list.append(take_action)
cube_1 = x[-1].copy()(list(action_map.keys())[pred[-1]])
new_list_sequences.append(x + [cube_1])
# print("new_list_sequences", len(new_list_sequences))
last_states_flat = [flatten_1d_b(x[-1]) for x in new_list_sequences]
value, _ = model.predict(np.array(last_states_flat), batch_size=1024)
value = value.ravel().tolist()
for x, v in zip(new_list_sequences, value):
x[-1].score = v if str(x[-1]) not in existing_cubes else -1
new_list_sequences.sort(key=lambda x: x[-1].score , reverse=True)
new_list_sequences = new_list_sequences[:100]
existing_cubes.update(set([str(x[-1]) for x in new_list_sequences]))
list_sequences = new_list_sequences
list_sequences.sort(key=lambda x: perc_solved_cube(x[-1]), reverse=True)
prec = perc_solved_cube((list_sequences[0][-1]))
if prec == 1:
success = True
break
return success, action_list
new_list_sequences.append(x + [cube_1])
new_list_sequences.append(x + [cube_2])
print("new_list_sequences", len(new_list_sequences))
last_states_flat = [flatten_1d_b(x[-1]) for x in new_list_sequences]
value, _ = model.predict(np.array(last_states_flat), batch_size=1024)
value = value.ravel().tolist()
for x, v in zip(new_list_sequences, value):
x[-1].score = v if str(x[-1]) not in existing_cubes else -1
new_list_sequences.sort(key=lambda x: x[-1].score, reverse=True)
new_list_sequences = new_list_sequences[:100]
existing_cubes.update(set([str(x[-1]) for x in new_list_sequences]))
list_sequences = new_list_sequences
list_sequences.sort(key=lambda x: perc_solved_cube(x[-1]),
reverse=True)
prec = perc_solved_cube((list_sequences[0][-1]))
print(prec)
if prec == 1:
break
print(perc_solved_cube(list_sequences[0][-1]))
print(list_sequences[0][-1])
last_states_flat = [flatten_1d_b(x[-1]) for x in new_list_sequences]
value, _ = model.predict(np.array(last_states_flat), batch_size=1024)
value = value.ravel().tolist()
for x, v in zip(new_list_sequences, value):
x[-1].score = v if str(x[-1]) not in existing_cubes else -1
new_list_sequences.sort(key=lambda x: x[-1].score , reverse=True)
new_list_sequences = new_list_sequences[:100]
existing_cubes.update(set([str(x[-1]) for x in new_list_sequences]))
list_sequences = new_list_sequences
list_sequences.sort(key=lambda x: perc_solved_cube(x[-1]), reverse=True)
#print(list_sequences[0])
preview_cube(take_action)
print([preview_cube])
prec = perc_solved_cube((list_sequences[0][-1]))
#print(prec)
if prec == 1:
break
#print(list_sequences[0])
#print(perc_solved_cube(list_sequences[0][-1]))
import utils #https://github.com/adrianliaw/PyCuber/blob/version2/examples/sample_program.py #To declare a cube object mycube = pc.Cube() st = utils.flatten_1d_b(mycube) print(st) # k=["R","S"] # #Formula is for set of actions taken # my_formula = pc.Formula(k[0]) # mycube(my_formula) # print(mycube) # #Reversing the actions # my_formula.reverse() # #A object for Algo alg = pc.Formula() #Random arrangement random_alg = alg.random() # random_alg = random_alg[:10] # print(random_alg) mycube(random_alg) print(utils.perc_solved_cube(mycube)) print(mycube)