def main(f_name, rot, snum):
sol = []
r = rubiks_cube()
with open("cnn_solver/" + f_name, 'w', newline = '') as file:
writer = csv.writer(file)
writer.writerow( ['state', 'move'] )
for i in range(0, snum):
state, sol, r = reverse_shuffle(rot)
while ( len(sol) != 0 ) :
s_ind = sol.pop(0)
s_ind_str = sh.convert_sol(s_ind, into_num = 1 )
state = sh.flatten_faces(r)
into = [ state, s_ind_str ]
writer.writerow( into )
sh.get_move( r, s_ind )()
state = sh.flatten_faces(r) # Gets the solved state
into = [ state, 12 ] # Inputs '0' for move on solved state
writer.writerow( into )
r.reset() # Reset cube for good measure
def test_obo(mod_name, num_shuf):
model = keras.models.load_model("cnn_solver/models/" + mod_name)
rc = rubiks_cube()
rc.shuffle(num_shuf)
i = 0
while (rc.if_solved() == 0):
rc_flat = sh.flatten_faces(rc) # into string
rc_np = sh.two_dim_data(rc_flat) # np array
rc_np = np.array(rc_np).reshape((1, 2, 12, 1)) # np reshape
out = model.predict_classes(rc_np)
out_p = sh.mv_num_to_char(int(out[0]))
print(out_p) # Print out prediction
mv = sh.get_move(rc, out[0]) # Get the callable fn move
mv() # Make the move
i += 1
if (i > 20):
print("No solve")
break
rc_flat = sh.flatten_faces(rc) # into string
print(rc_flat)
def solve_back(sol, rc): for i in range(0, len(sol)): temp = sh.get_move(rc, sol[i]) temp() print(sh.flatten_faces(rc)) return rc
def main():
#create rubiks_cube object
cube = rubiks_cube()
mod_name = input("What model do you want to use? ")
shuf_num = input("How many shuffles? ")
cube.shuffle(int(shuf_num))
model = keras.models.load_model("cnn_solver/models/" + mod_name)
last_move = -1 #keeps track of last move done
rand_check = 0 #makes sure we don't keep calling random continuously
repeat_same_move = 0 #keeps track of repeating the same move
count = 0 #keeps track of all moves done
mvs = names_of_moves() #used for conversion to letter for comparison
while(cube.if_solved() == 0) :
#model reads in data and makes prediction
flat = flatten_faces(cube)
data = two_dim_data(flat)
out = model.predict_classes(data.reshape(1,2,12,1))
#if this move is the counter of the last move (model is repeating itself) or if model has repeated same move 4 times
if((mvs[last_move] == counter_move(mvs[out[0]]) and rand_check != 1 and last_move != -1) or (repeat_same_move == 4)):
#perform a random move on cube
ranm = random_move(cube)
ranm_out = mv_num_to_char(ranm)
p_out = mv_num_to_char(int(out[0]))
print(p_out),
print("Called random move:", ranm_out)
#rand_check makes sure we don't call random check multiple times in a row
rand_check = 1
#reset counter for repeated moves
repeat_same_move = 0
else:
#perform model's predicted move
p_out = mv_num_to_char(int(out[0]))
print(p_out)
mv = get_move(cube, out[0])
mv()
#reset counter for repeated moves to 0 if the last move was different than current move
if(out[0] != last_move) :
repeat_same_move = 0
else:
repeat_same_move+=1
#keep track of last move
last_move = out[0]
rand_check = 0
#while loop ends if we have done 50 moves
count+=1
if(count == 50) :
print("No solve (in less than 50 moves)")
break
final_state = flatten_faces(cube)
print(final_state)
def ml_test_random(model_name, f_name, ub=20, each=100, max_allow=20):
model = keras.models.load_model("cnn_solver/models/" + model_name)
rc = rubiks_cube()
mvs = sh.names_of_moves() # used for conversion to letter for comparison
with open(f_name, 'w', newline='') as file:
writer = csv.writer(file)
writer.writerow(['rotations', 'solve_prop', 'solve_rotations'])
for i in range(1, ub + 1): # 1 - ub rotations
solve_sum = 0 # Reset sum variables before another iteration
rot_sum = 0
for k in range(0, each): # Counts the number of times
rc.shuffle(i)
solve = 0 # Checks if we have a solve or not
j = 0 # Keeps track of all moves done
last_move = -1 # Keeps track of last move done
rand_check = 0 # Makes sure we don't keep calling random
repeat_same_move = 0 # Keeps track of repeating the same move
while (rc.if_solved() == 0):
rc_flat = sh.flatten_faces(rc) # into string
rc_np = sh.two_dim_data(rc_flat) # np array
rc_np = np.array(rc_np).reshape(
(1, 2, 12, 1)) # np reshape
out = model.predict_classes(rc_np) # prediction from model
if (out[0] == 12):
if (rc.if_solved() == 1):
solve = 1
break
else:
solve = 0
break
if ((mvs[last_move] == sh.counter_move(mvs[out[0]])
and rand_check != 1 and last_move != -1)
or (repeat_same_move == 4)):
ranm = random_move(cube) # Calls random move
rand_check = 1
repeat_same_move = 0
else:
if (out[0] != 12):
#Get the callable function move
mv = sh.get_move(rc, out[0])
mv() # Make the move
if (out[0] !=
last_move): # Reset counter if different move
repeat_same_move = 0
else:
repeat_same_move += 1
j += 1
# If we solve the cube
if (rc.if_solved() == 1):
solve = 1
break
# If we reach the maximum number of rotations allowed for the model
if (j > max_allow):
solve = 0
break
if (solve == 1): # If the model solved the cube
solve_sum += 1
rot_sum += j
rc.reset() # Reset the cube before next trial
progress = 100 * ((k + (each * (i - 1))) / (each * ub))
print("Progres:", progress, "%")
solve_prop = (solve_sum) / (
each) # Proportion of times the model solved the cube
if (solve_sum > 0):
rot_prop = (rot_sum) / (
solve_sum) # Number of rotations per each solve trial
else:
rot_prop = 0
r = [i, solve_prop, rot_prop] # Input into the csv file
writer.writerow(r) # Write to csv file
def ml_test(model_name, f_name, ub=20, each=100, max_allow=20):
model = keras.models.load_model("cnn_solver/models/" + model_name)
rc = rubiks_cube()
with open(f_name, 'w', newline='') as file:
writer = csv.writer(file)
writer.writerow(['rotations', 'solve_prop', 'solve_rotations'])
for i in range(1, ub + 1): # 1 - ub rotations
solve_sum = 0 # Reset sum variables before another iteration
rot_sum = 0
for k in range(0, each): # Counts the number of times
rc.shuffle(i)
solve = 0
j = 0
while (rc.if_solved() == 0):
# If we reach the maximum number of rotations allowed for the model
if (j > max_allow):
solve = 0
break
rc_flat = sh.flatten_faces(rc) # into string
rc_np = sh.two_dim_data(rc_flat) # np array
rc_np = np.array(rc_np).reshape(
(1, 2, 12, 1)) # np reshape
out = model.predict_classes(rc_np) # prediction from model
if (out[0] != 12):
mv = sh.get_move(rc,
out[0]) # Get the callable fn move
mv() # Make the move
j += 1
# If we solve the cube
if (rc.if_solved() == 1):
solve = 1
break
if (solve == 1): # If the model solved the cube
solve_sum += 1
rot_sum += j
rc.reset() # Reset the cube before next trial
progress = 100 * ((k + (each * (i - 1))) / (each * ub))
print("Progres:", progress, "%")
solve_prop = (solve_sum) / (
each) # Proportion of times the model solved the cube
if (solve_sum > 0):
rot_prop = (rot_sum) / (
solve_sum) # Number of rotations per each solve trial
else:
rot_prop = 0
r = [i, solve_prop, rot_prop] # Input into the csv file
writer.writerow(r) # Write to csv file