def neh_plus(N, m):
k = 1
W = [[sum(x), x] for x in N]
pi = []
pi_star = []
W.sort(key=lambda x: x[0])
while len(W) != 0:
j = W[len(W) - 1][1]
pi_star.append(j)
for l in range(k):
pi.insert(l, j)
if FSPService.loss_function(pi, m) < FSPService.loss_function(pi_star, m):
pi_star = cp.copy(pi)
del pi[l]
pi = cp.copy(pi_star)
del W[len(W) - 1]
if k > 1:
x = NEH.select_x(pi, m, j)
pi.remove(x)
for l in range(k):
pi.insert(l, x)
if FSPService.loss_function(pi, m) < FSPService.loss_function(pi_star, m):
pi_star = cp.copy(pi)
del pi[l]
pi = cp.copy(pi_star)
k += 1
return FSPService.loss_function(pi_star, m), pi_star
def bound_1(Pi, N, m, data) -> int:
max_val = -1
E = [sum(col) for col in zip(*N)]
for i in range(1, m + 1):
Ci = service.loss_function(Pi, i)
if Ci + E[i - 1] > max_val:
max_val = Ci + E[i - 1]
return max_val
def neh(N, m):
k = 1
W = [[sum(x), x] for x in N]
pi = []
pi_star = []
W.sort(key=lambda x: x[0])
while len(W) != 0:
j = W[len(W) - 1][1]
pi_star.append(j)
for l in range(0, k):
pi.insert(l, j)
if FSPService.loss_function(pi, m) < FSPService.loss_function(pi_star, m):
pi_star = cp.copy(pi)
del pi[l]
pi = cp.copy(pi_star)
k += 1
del W[len(W) - 1]
return FSPService.loss_function(pi_star, m), pi
def recursive(S, C_max, leaf, best, m):
if len(S) > 0:
for i in S:
s = copy.copy(S)
s.remove(i)
l = copy.copy(leaf)
l.append(i)
recursive(s, C_max, l, best, m)
else:
C_tmp = FSPService.loss_function(leaf, m)
if C_max[0] > C_tmp:
C_max[0] = C_tmp
best[0] = leaf
def select_x(pi: [], m, j):
x = None
C_max = math.inf
for i in range(len(pi)):
if pi[i] == j:
continue
tmp_task = pi[i]
del pi[i]
tmp_C_max = FSPService.loss_function(pi, m)
if tmp_C_max < C_max:
C_max = tmp_C_max
x = tmp_task
pi.insert(i, tmp_task)
return x
def procedure(J, n, m, L, T, reduce_temp_fcn, arg_reduce_temp,
init_solution_fcn, move_fcn):
T_end = 0.05
pi_new = None
pi_star = init_solution_fcn(J, m)
pi = init_solution_fcn(J, m)
while T > T_end:
for k in range(L):
i = random.randint(0, n - 1)
j = random.randint(0, n - 1)
pi_new = move_fcn(pi_star, i, j)
delta_C_max = FSPService.loss_function(
pi, m) - FSPService.loss_function(pi_new, m)
if delta_C_max > 0:
r = random.uniform(0, 1)
if r >= np.e**(delta_C_max / T):
pi_new = cp.copy(pi)
pi = pi_new
if FSPService.loss_function(pi, m) < FSPService.loss_function(
pi_star, m):
pi_star = cp.copy(pi)
T = reduce_temp_fcn(T, arg_reduce_temp)
print(FSPService.loss_function(pi, m))
def BnB(j, N_, Pi_):
Pi = copy.deepcopy(Pi_)
N = copy.deepcopy(N_)
Pi.append(j)
N.remove(j)
if len(N) > 0:
LB = lb(Pi, N, m, data)
pass
if LB <= UB[0]:
for i in N:
BnB(i, N, Pi)
else:
Cmax = service.loss_function(Pi, m)
if Cmax < UB[0]:
UB[0] = Cmax
Pi_star[0] = Pi
def bound_3(Pi, N, m, data):
max_val = -1
E = [sum(col) for col in zip(*N)]
for i in range(1, m + 1):
Ci = service.loss_function(Pi, i)
sum_k = 0
for k in range(i, m):
min_val = math.inf
for j in range(len(N)):
if N[j][k] < min_val:
min_val = N[j][k]
sum_k += min_val
if Ci + E[i - 1] + sum_k > max_val:
max_val = Ci + E[i - 1] + sum_k
return max_val
def bound_4(Pi, N, m, data):
max_val = -1
E = [sum(col) for col in zip(*N)]
for i in range(m):
Ci = service.loss_function(Pi, i + 1)
min_val = math.inf
# sum_k = None
for j in range(len(N)):
sum_k = 0
for k in range(i + 1, m):
sum_k += N[j][k]
min_val = min(min_val, sum_k)
if Ci + E[i] + min_val > max_val:
max_val = Ci + E[i] + min_val
return max_val
def procedure(self):
population = self.generate_start_population(self.n)
for j in range(self.epoch):
new_population = [] + cp.copy(population)
parents = Genetic.selection(population)
# selection
for pair in parents:
child = Genetic.crossover(pair[1][0], pair[0][0])
new_population.append([child, self.calc_value(child)])
# crossover
random_mutation_quantity = random.randint(1, 5)
for i in range(random_mutation_quantity):
mutant = Genetic.mutation(random.choice(population)[0])
new_population.append([mutant, self.calc_value(mutant)])
new_population.sort(key=lambda x: -x[1])
population = new_population[:len(population)]
return FSPService.loss_function(population[0][0],
self.m), population[0][0]
def ub_2(data, m):
return service.loss_function(data, m)
def calc_value(self, J):
return 1 / (FSPService.loss_function(J, self.m) / self.normalizer)
def __init__(self, path_to_data, epoch):
self.n, self.m, self.data = NEHService.read_data_by_rows(path_to_data)
self.epoch = epoch
self.normalizer = FSPService.loss_function(self.data, self.m) / 2
return T - x
@staticmethod
def reduce_temperature_geometric(T, alpha):
return T * alpha
@staticmethod
def reduce_temperature_log(T, it):
b = it + 1
return T / np.log(b)
if __name__ == '__main__':
filepath = 'D:\Programming\python\SPD\\neh\dane\\ta003.txt'
n, m, data = NEHService.read_data_by_rows(filepath)
natural_c_max = FSPService.loss_function(data, m)
print(f'natural Cmax = {natural_c_max}')
L = [10**2, 10**3, 10**4]
T_0 = [math.sqrt(n), n, n**2]
temp_reduce_fcn_arr = [
SimulatedAnnealing.reduce_temperature_linear,
SimulatedAnnealing.reduce_temperature_geometric,
SimulatedAnnealing.reduce_temperature_log
]
x = [T_0[0] / (10**3), T_0[0] / (10**4), T_0[0] / (10**5)]
alpha = [.97, .95, .9]
init_solution_fcn_arr = [
SimulatedAnnealing.init_solution_1, SimulatedAnnealing.init_solution_2,
SimulatedAnnealing.init_solution_3
]
move_fcn_arr = [