def neh(tasks, numb_of_machines):
start = timer()
# step 1: find omegas(j)
omegas = []
for task in tasks:
omegas.append(sum(task.times))
# step 2: sort in descending order (get sorted order)
omegas_order = np.argsort(-np.array(omegas)).tolist()
# steps 3, 4: repeat n times (n = numb of tasks)
solution_order = []
for i in omegas_order: # (3) get task with the highest omega value
# (4) insert task & pick task with the lowest makespan
lowest_makespan = float("inf")
lowest_makespan_sequence = []
sequences = get_sequences(i, solution_order)
for sequence in sequences:
if makespan(sequence, tasks, numb_of_machines) < lowest_makespan:
lowest_makespan = makespan(sequence, tasks, numb_of_machines)
lowest_makespan_sequence = sequence
solution_order = lowest_makespan_sequence
#print(lowest_makespan)
stop = timer()
return solution_order, (stop - start) * 1000
def get_probability(prev_order, new_order, tasks, temperature):
prev_order_cmax = makespan(prev_order, tasks)
new_order_cmax = makespan(new_order, tasks)
if new_order_cmax < prev_order_cmax:
return 1
return exp((prev_order_cmax - new_order_cmax) / temperature)
def generating(iter):
number_of_machines = int(input("Ile maszyn?: "))
number_of_tasks = int(input("Ile zadan?: "))
times = []
durationBruteforce = []
durationJohnson = []
i = 0
while i < iter:
generatedTasks = []
cnt_tasks = 0
cnt_machines = 0
for cnt_tasks in range(0, number_of_tasks):
rows = []
for cnt_machines in range(0, number_of_machines):
rows.append(int(random.uniform(1, 10)))
print("{}".format(rows))
generatedTasks.append(Task(cnt_tasks, rows))
bruteforceOrder, timeBruteforce = bruteforce(
copy.deepcopy(generatedTasks), number_of_machines)
johnsonOrder, timeJohnson = johnson(copy.deepcopy(generatedTasks),
number_of_machines)
durationBruteforce.append(timeBruteforce)
durationJohnson.append(timeJohnson)
bruteforceMakespan = makespan(bruteforceOrder, generatedTasks,
number_of_machines)
johnsonMakespan = makespan(johnsonOrder, generatedTasks,
number_of_machines)
i += 1
if johnsonMakespan == bruteforceMakespan:
times.append(johnsonMakespan)
else:
times.append(-1)
x = PrettyTable()
print("")
print("----------------------------------------------------------")
x.field_names = [
"l.p.", "Johnson Makespan", "Poprawnosc", "Czas Bruteforce [ms]",
"Czas Johnson [ms]"
]
k = 0
for k in range(0, iter):
if times[k] == -1:
x.add_row([
k + 1, "{}".format(times[k]), "Nie",
"{}".format(durationBruteforce[k]),
"{}".format(durationJohnson[k])
])
else:
x.add_row([
k + 1, "{}".format(times[k]), "Tak",
"{}".format(durationBruteforce[k]),
"{}".format(durationJohnson[k])
])
print(x)
def IR4_mod(tasks, solution_order, numb_of_machines):
maxDiff = 0
baseTime = makespan(solution_order, tasks, numb_of_machines)
for i in solution_order:
idx = solution_order.index(i)
solution_order.remove(i)
tempTime = makespan(solution_order, tasks, numb_of_machines)
if baseTime - tempTime > maxDiff:
maxDiff = baseTime - tempTime
maxTask = i
solution_order.insert(idx, i)
return maxTask
def amelioration_locale(individu, data):
qualite_initiale = makespan(individu, data)
voisins = []
for _ in range(5):
voisin = swap(individu)
voisins.append((voisin, makespan(voisin, data)))
meilleurVoisin = choisir_meilleur(voisins)
if (meilleurVoisin[1] < qualite_initiale):
individu = meilleurVoisin[0]
return individu
def neh(jobMatrix):
jobs_with_total_times = [(job_id, sum(job))
for job_id, job in enumerate(jobMatrix.T)]
order = []
for job in sorted(jobs_with_total_times, key=lambda x: x[1], reverse=True):
candidates = []
for i in range(0, len(order) + 1):
candidate = order[:i] + [job[0]] + order[i:]
candidates.append((candidate, makespan(candidate, jobMatrix)))
order = min(candidates, key=lambda x: x[1])[0]
#print(makespan(order, jobMatrix))
return (order, makespan(order, jobMatrix))
def bruteforce(tasks, numb_of_machines):
start = timer()
best_order = get_order(tasks)
best_makespan = makespan(best_order, tasks, numb_of_machines)
for p in permute(get_order(tasks)):
print("order: {}".format(p))
print("makespan: {}".format(makespan(p, tasks, numb_of_machines)))
print("---")
if makespan(p, tasks, numb_of_machines) < best_makespan:
best_order = list(p)
best_makespan = makespan(p, tasks, numb_of_machines)
stop = timer()
return best_order, (stop - start) * 1000
def plotGantt(jobMatrix, jobOrder, nom, nb=7):
fig, ax = plt.subplots()
nb_machine, nb_jobs = jobMatrix.shape
ganttTable = _calc_makespan(jobMatrix, jobOrder, True)
colors = [
'#e6194b', '#3cb44b', '#ffe119', '#4363d8', '#f58231', '#911eb4',
'#46f0f0', '#f032e6', '#bcf60c', '#fabebe', '#008080', '#e6beff',
'#9a6324', '#fffac8', '#800000', '#aaffc3', '#808000', '#ffd8b1',
'#000075', '#808080', '#ffffff', '#000000', "#ff0000", "#7f0000",
"#400000", "#ff8080", "#7f4040", "#403030", "#bf8f00", "#7f7040",
"#7fff00", "#508020", "#dfffbf", "#10401c", "#7fff9f", "#208080",
"#80ffff", "#204040", "#002080", "#101c40", "#809fff", "#bfcfff",
"#9f40ff", "#604080", "#dfbfff", "#383040", "#400030", "#bf309b",
"#806078"
]
for i in range(nb_jobs):
ax.broken_barh([(ganttTable[i, 2 * j],
ganttTable[i, 2 * j + 1] - ganttTable[i, 2 * j])
for j in range(nb_machine)], (10 * i, 10),
facecolors=colors[:nb_machine])
ax.set_xlabel('Temps')
ax.set_yticks([i * 10 for i in range(nb)])
tasklist = ["Task" + str(x) for x in jobOrder]
ax.set_yticklabels(tasklist)
ax.grid(True)
plt.title('Makespan = {}'.format(makespan(jobOrder, jobMatrix)))
plt.show()
def test(file, testcase):
start = time()
result = makespan(file(testfile), testfile)
end = time()
print "Best Makespan =", result
return (result, end - start)
def random_search(tasks, max_time):
numb_of_experiments = 100 # experiments performed per loop
best_solution = ""
best_cmax = 10000000
t0 = time.time()
total_experiments = 0
rs = get_random_schedule(tasks)
while True:
start = time.time()
for i in range(0, numb_of_experiments):
random.shuffle(rs)
cmax = makespan(rs, tasks)
if cmax < best_cmax:
best_cmax = cmax
best_solution = copy.deepcopy(rs)
total_experiments += numb_of_experiments
if max_time and time.time() - t0 > max_time:
break
t = time.time() - start
#if t > 0:
#print("Best: ", best_cmax, ", time: ", time.time() - t0)
return best_solution
def neh(times):
jobs_with_total_times = [(job_id, sum(job)) for job_id, job in enumerate(times)]
order = []
for job in sorted(jobs_with_total_times, key=lambda x: x[1], reverse = True):
candidates = []
for i in range(0, len(order) + 1):
candidate = order[:i] + [job[0]] + order[i:]
candidates.append((candidate, makespan(candidate, times)))
order = min(candidates, key = lambda x: x[1])[0]
return order
def get_probability_sa(prev_order, new_order, tasks, numb_of_machines, temperature, move_type):
prev_order_cmax = makespan(prev_order, tasks, numb_of_machines)
new_order_cmax = makespan(new_order, tasks, numb_of_machines)
if move_type == 0:
if new_order_cmax < prev_order_cmax:
return 1
return 1/(1+2*exp((prev_order_cmax-new_order_cmax)/temperature))
elif move_type == 1:
if new_order_cmax < prev_order_cmax:
return 0
return 1/(1+2*exp((prev_order_cmax-new_order_cmax)/temperature))
elif move_type == 2:
if prev_order_cmax != new_order_cmax:
if new_order_cmax < prev_order_cmax:
return 1
return 1/(1+2*exp((prev_order_cmax-new_order_cmax)/temperature))
else:
return 0
return -1
def cds(times):
jobs_count = len(times)
machine_count = len(times[0])
perms = []
times_merged = [[0, sum(job_times)] for job_times in times]
for i in range(0, machine_count - 1):
for k in range(0, jobs_count):
times_merged[k][0] += times[k][i]
times_merged[k][1] -= times[k][i]
perms.append(johnson(times_merged))
return min(perms, key=lambda p: makespan(p, times))
def cds(times):
jobs_count = len(times)
machine_count = len(times[0])
perms = []
times_merged = [[0, sum(job_times)] for job_times in times]
for i in range(0, machine_count - 1):
for k in range(0, jobs_count):
times_merged[k][0] += times[k][i]
times_merged[k][1] -= times[k][i]
perms.append(johnson(times_merged))
return min(perms, key=lambda p: makespan(p, times))
def CDS(jobMatrix):
times = jobMatrix.T
jobs_count = len(times)
machine_count = len(times[0])
merged_times = [[0, sum(j)] for j in times]
perms = []
for i in range(0, machine_count - 1):
for j in range(0, jobs_count):
merged_times[j][0] += times[j][i]
merged_times[j][1] -= times[j][i]
perms.append(Johnson(np.array(merged_times).T))
result = min(perms, key=lambda p: _calc_makespan(jobMatrix, np.array(p)))
return (result, makespan(result, jobMatrix))
def do_schrage(tasks):
ready_tasks = []
final_list = []
unready_tasks = copy.deepcopy(tasks)
t = min(extract_column(unready_tasks, 0))
while ready_tasks or unready_tasks:
while unready_tasks and min(extract_column(unready_tasks, 0)) <= t:
j = numpy.argmin(extract_column(unready_tasks, 0))
ready_tasks.append(unready_tasks.pop(j))
if not ready_tasks:
t = min(extract_column(unready_tasks, 0))
else:
j = numpy.argmax(extract_column(ready_tasks, 2))
final_list.append(ready_tasks.pop(j))
t += final_list[-1].times[1]
return makespan.makespan(final_list), final_list
def do_carlier(carlier_object):
carlier_object.U, carlier_object.pi = schrage.do_schrage(
carlier_object.tasks)
if carlier_object.U < carlier_object.UB:
carlier_object.UB = copy.deepcopy(carlier_object.U)
carlier_object.opt_pi = copy.deepcopy(carlier_object.pi)
# b index
potential_b = []
for j in range(0, len(carlier_object.tasks)):
if makespan.makespan(carlier_object.pi) == (
carlier_object.pi[j].times[2] +
makespan.makespan_carlier(carlier_object.pi)[j]):
potential_b.append(carlier_object.pi[j].id)
carlier_object.b = max(potential_b)
# a index
for j in range(0, len(carlier_object.tasks)):
sum_of = 0
for i in range(j, find_by_id(carlier_object, carlier_object.b) + 1):
sum_of += carlier_object.pi[i].times[1]
if carlier_object.U == (
sum_of + carlier_object.pi[j].times[0] +
carlier_object.tasks[carlier_object.b].times[2]):
carlier_object.a = carlier_object.pi[j].id
break
carlier_object.a = find_by_id(carlier_object, carlier_object.a)
carlier_object.b = find_by_id(carlier_object, carlier_object.b)
# c index
potential_c = []
for j in range(carlier_object.a, carlier_object.b + 1):
if carlier_object.pi[j].times[2] < carlier_object.pi[
carlier_object.b].times[2]:
potential_c.append(j)
if potential_c:
carlier_object.c = max(potential_c)
else:
return carlier_object.U
# K block
K = [] # K to blok, w którym są indeksy do pi[index]
for j in range(carlier_object.c + 1, carlier_object.b + 1):
K.append(j)
# K u {c} block
K_ = [] # K to blok, w którym są indeksy do pi[index]
for j in range(carlier_object.c, carlier_object.b + 1):
K_.append(j)
# r(K), q(K), p(K), h(K)
h_k, r_k, p_k, q_k = calc_k(carlier_object, K)
# remember tasks[c].times[0]
old_r_pi = copy.deepcopy(
carlier_object.tasks[carlier_object.pi[carlier_object.c].id].times[0])
# new tasks[c].times[0]
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[0] = max(
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[0], r_k + p_k)
carlier_object.LB = schrage.do_schrage_pmtn(carlier_object.tasks)[0]
# step 16 new LB
h_k_c, r_k_c, p_k_c, q_k_c = calc_k(carlier_object, K_)
carlier_object.LB = max(h_k, h_k_c, carlier_object.LB)
# left if
if carlier_object.LB < carlier_object.UB:
do_carlier(carlier_object)
# restore tasks[c].times[0]
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[0] = old_r_pi
# remember tasks[c].times[2]
old_q_pi = copy.deepcopy(carlier_object.pi[carlier_object.c].times[2])
# new q time for tasks[c].times[2]
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[2] = max(
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[2], q_k + p_k)
# step 22 new LB
carlier_object.LB = schrage.do_schrage_pmtn(carlier_object.tasks)[0]
# step 23 max LB
h_k_c, r_k_c, p_k_c, q_k_c = calc_k(carlier_object, K_)
carlier_object.LB = max(h_k, h_k_c, carlier_object.LB)
# step 24 right loop
if carlier_object.LB < carlier_object.UB:
do_carlier(carlier_object)
# step 27 restore tasks[c].times[2]
carlier_object.tasks[carlier_object.pi[
carlier_object.c].id].times[2] = old_q_pi
return carlier_object.U
import copy
from datareader import get_data
from makespan import makespan, to_natural_order, get_order
from schrage import schrage_n2, schrage_n2_pmtn, schrage_nlogn, schrage_nlogn_pmtn
from random_search import random_search
tasks = get_data("in50.txt")
# INITIAL ORDER
init_order = get_order(tasks)
init_makespan = makespan(init_order, tasks)
print("[INIT] makespan: ", init_makespan)
# SCHRAGE ORDER
schrage_n2_order, schrage_n2_time = schrage_n2(tasks)
shrage_n2_makespan = makespan(schrage_n2_order, tasks)
#print("[SHRAGE N^2] order: ", schrage_n2_order)
print("[SHRAGE N^2] makespan: {}, time: {}" .format(shrage_n2_makespan, schrage_n2_time))
# SCHRAGE ORDER NLOGN
schrage_nlogn_order, schrage_nlogn_time = schrage_nlogn(tasks)
schrage_nlogn_makespan = makespan(schrage_nlogn_order, tasks)
#print("[SHRAGE NLOGN] order: ", schrage_nlogn_order)
print("[SHRAGE NLOGN] makespan: {}, time: {}" .format(schrage_nlogn_makespan, schrage_nlogn_time))
#SCHRAGE ORDER N2 PMTN
schrage_n2_ptmn_makespan, schrage_n2_ptmn_order, schrage_n2_ptmn_time = schrage_n2_pmtn(tasks)
print("[SHRAGE N^2 PMTN] makespan: {}, time: {}" .format(schrage_n2_ptmn_makespan, schrage_n2_ptmn_time))
#SCHRAGE ORDER NLOGN PMTN
schrage_nlogn_pmtn_makespan, schrage_nlogn_pmtn_order, schrage_nlogn_pmtn_time = schrage_nlogn_pmtn(tasks)
fcn0_cmax = []
fcn1_times = []
fcn1_cmax = []
move_type = 0
# for FCN0
for set in sets:
tasks, numb_of_machines = get_data(set)
cooling_fcn_type = 0
simulated_annealing_order, iterations, sa_time = simulated_annealing(copy.deepcopy(tasks), numb_of_machines,
init_temp, final_temp, u, cooling_fcn_type,
move_type, insert)
simulated_annealing_makespan = makespan(simulated_annealing_order, tasks, numb_of_machines)
fcn0_times.append(sa_time)
fcn0_cmax.append(simulated_annealing_makespan)
# for FCN1
for set in sets:
tasks, numb_of_machines = get_data(set)
cooling_fcn_type = 1
simulated_annealing_order, iterations, sa_time = simulated_annealing(copy.deepcopy(tasks), numb_of_machines,
init_temp, final_temp, u, cooling_fcn_type,
move_type, insert)
simulated_annealing_makespan = makespan(simulated_annealing_order, tasks, numb_of_machines)
fcn1_times.append(sa_time)
fcn1_cmax.append(simulated_annealing_makespan)
def test(f, testcase):
start = time()
result = makespan(f(testcase), testcase)
end = time()
return (result, end - start)
from prettytable import PrettyTable as tabelki
import makespan
import read
import schrage
x = read.Reader()
x.read("data//in50.txt")
y = schrage.Schrage(x.my_data)
y_1 = schrage.Schrage(x.my_data)
t = tabelki(['Alogrithm', 'in50', 'in100', 'in200', 'Sum'])
sigma = y.do_schrage()
c_max = makespan.makespan(sigma)
c_max_pmtn = y_1.do_schrage_pmtn()
x.read("data//in100.txt")
y_2 = schrage.Schrage(x.my_data)
y_3 = schrage.Schrage(x.my_data)
sigma = y_2.do_schrage()
c_max_1 = makespan.makespan(sigma)
c_max_pmtn_1 = y_3.do_schrage_pmtn()
x.read("data//in200.txt")
y_4 = schrage.Schrage(x.my_data)
y_5 = schrage.Schrage(x.my_data)
sigma = y_4.do_schrage()
c_max_2 = makespan.makespan(sigma)
def evaluer_qualite(population, data):
return [(individual, makespan(individual, data))
for individual in population]
taskset = []
x = []
for i in range(5, 300):
taskset.append(create_instances(i))
x.append(i)
iter = 0
for i in range(0, len(taskset)):
tasks = copy.deepcopy(taskset[i])
# ------------------------------------------------ SCHRAGE
# SCHRAGE ORDER
schrage_n2_order, schrage_n2_time = schrage_n2(tasks)
shrage_n2_makespan = makespan(schrage_n2_order, tasks)
# print("[SHRAGE N^2] order: ", schrage_n2_order)
schrage_makespans.append(shrage_n2_makespan)
schrage_times.append(schrage_n2_time)
# ------------------------------------------------ SCHRAGE
# SCHRAGE ORDER N2 PMTN
schrage_n2_ptmn_makespan, schrage_n2_ptmn_order, schrage_n2_ptmn_time = schrage_n2_pmtn(
tasks)
schrage_pmtn_makespans.append(schrage_n2_ptmn_makespan)
schrage_pmtn_times.append(schrage_n2_ptmn_time)
# ------------------------------------------------ DEEP LEFT
ub = 999999999
u, pi = schrage(copy.deepcopy(tasks))
def optimal(times):
job_count = len(times)
return min(permutations(range(job_count)),
key=lambda x: makespan(x, times))
def test(f, testcase):
start = time()
result = makespan(f(testcase), testcase)
end = time()
return (result, end - start)
def optimal(times):
job_count = len(times)
return min(permutations(range(job_count)), key = lambda x: makespan(x, times))
def evaluate_fitness(population, times):
return [(individual, makespan(individual, times))
for individual in population]
def simulated_annealing(matrice, Ti = 790,Tf = 3 ,alpha = 0.93):
#Number of jobs given
nb_machines, job_count = matrice.shape
n = job_count;
default_timer = None
if sys.platform == "win32":
default_timer = time.time()
else:
default_timer = time.time()
s = default_timer
#Initialize the primary seq
old_seq = neh(matrice)
old_seq = old_seq[0]
old_makeSpan = makespan(old_seq,matrice)
#print("old sequence: ",old_seq)
#print("old makespan: ",old_makeSpan)
new_seq = []
delta_mk1 = 0
#Initialize the temperature
T = Ti
Tf = Tf
alpha = alpha
# of iterations
temp_cycle = 0
while T >= Tf :
new_seq = old_seq.copy()
job = new_seq.pop(randint(0,n-1))
new_seq.insert(randint(0,n-1),job)
new_make_span = makespan(new_seq,matrice)
delta_mk1 = new_make_span - old_makeSpan
if delta_mk1 <= 0:
old_seq = new_seq
old_makeSpan = new_make_span
else :
Aprob = np.exp(-(delta_mk1/T))
if Aprob > np.random.uniform(0.5,0.9):
old_seq = new_seq
old_makeSpan = new_make_span
else :
#The solution is discarded
pass
T = T * alpha
temp_cycle += 1
e = default_timer
#Result Sequence
seq = old_seq
schedules = np.zeros((nb_machines, job_count), dtype=dict)
# schedule first job alone first
task = {"name": "job_{}".format(
seq[0] + 1), "start_time": 0, "end_time": matrice[0][seq[0]]}
schedules[0][0] = task
for m_id in range(1, nb_machines):
start_t = schedules[m_id - 1][0]["end_time"]
end_t = start_t + matrice[m_id][0]
task = {"name": "job_{}".format(
seq[0] + 1), "start_time": start_t, "end_time": end_t}
schedules[m_id][0] = task
for index, job_id in enumerate(seq[1::]):
start_t = schedules[0][index]["end_time"]
end_t = start_t + matrice[0][job_id]
task = {"name": "job_{}".format(
job_id + 1), "start_time": start_t, "end_time": end_t}
schedules[0][index + 1] = task
for m_id in range(1, nb_machines):
start_t = max(schedules[m_id][index]["end_time"],
schedules[m_id - 1][index + 1]["end_time"])
end_t = start_t +matrice[m_id][job_id]
task = {"name": "job_{}".format(
job_id + 1), "start_time": start_t, "end_time": end_t}
schedules[m_id][index + 1] = task
t_t = e - s
return seq, old_makeSpan
import copy
from datareader import get_data
from makespan import makespan, to_natural_order, get_order
from simulated_annealing import simulated_annealing
from improved_simulated_annealing import improved_simulated_annealing
from neh import neh
tasks, numb_of_machines = get_data("data.001")
# INITIAL ORDER
init_order = get_order(tasks)
init_makespan = makespan(init_order, tasks, numb_of_machines)
print("[INIT] makespan: {}, time: {}" .format(init_makespan, 0))
# NEH ORDER
neh_order, neh_time = neh(copy.deepcopy(tasks), numb_of_machines)
neh_makespan = makespan(neh_order, tasks, numb_of_machines)
print("[NEH ] makespan: {}, time: {}" .format(neh_makespan, neh_time))
# SIMULATED ANNEALING ORDER
init_temp = 5000
final_temp = 0.1
u = 0.98
cooling_fcn_type = 0
move_type = 0
insert = 0
simulated_annealing_order, iterations_sa, sa_time = simulated_annealing(copy.deepcopy(tasks), numb_of_machines, init_temp,
final_temp, u, cooling_fcn_type, move_type, insert)
simulated_annealing_makespan = makespan(simulated_annealing_order, tasks, numb_of_machines)
def evaluate_fitness(population, times):
return [(individual, makespan(individual, times)) for individual in population]
def opt(times):
job_count = len(times)
print "Machine times: \n", times, '\n'
z = min(permutations(range(job_count)), key=lambda x: makespan(x, times))
return z
