def optimize_ga(timetable, func):
sorter = GeneticSorter(timetable, func, seed=(timetable,), population_size=100, crossover_p=0.3, mutations={
inversion_mutation: 0.002,
swap_mutation: 0.02,
displacement_mutation: 0.01,
})
return sorter.run(100)
def optimize_ga(timetable, func, params):
params = params or {}
population_size = params.get('population_size', 100)
sorter = GeneticSorter(timetable, func, seed=(timetable,), population_size=population_size, crossover_p=0.2, mutations={
inversion_mutation: 0.005,
swap_mutation: 0.05,
displacement_mutation: 0.03,
random_replacement_mutation: 0.01,
})
return sorter.run(params.get('iterations', 100))
def test_list_sort(self):
N = 20
data = random.sample(xrange(N), N)
f = lambda p: sum(1.0 / (abs(i-x) + 1) for i, x in enumerate(p))
sorter = GeneticSorter(data, evaluation_func=f, population_size=50, mutations={swap_mutation: 0.02})
# This is fragile: if we don't find a solution in 10^6 generations this test will fail.
for i in xrange(1000):
result = sorter.run(1000)
if result == tuple(xrange(N)):
break
self.failUnlessEqual(result, tuple(xrange(N)))
def test_list_sort(self):
'''Tests if the GeneticSorter correctly sorts a supplied list.
'''
N = 20
data = random.sample(range(N), N)
f = lambda p: sum(1.0 / (abs(i-x) + 1) for i, x in enumerate(p))
sorter = GeneticSorter(data, evaluation_func=f, population_size=50, mutations={swap_mutation: 0.02})
# This is fragile: if we don't find a solution in 10^6 generations this test will fail.
for i in range(1000):
result = sorter.run(1000)
if result == tuple(range(N)):
break
self.assertEqual(result, tuple(range(N)))
def handle(self, **options):
submission_data = {}
meeting = Meeting.objects.create(start=datetime.datetime(
2010, 4, 13, 10),
title="Test")
entries_by_title = {}
for name, submissions, comments in csv.reader(open(
os.path.join(os.path.dirname(__file__), 'data',
'meeting100413.csv'), 'r'),
delimiter=',',
quotechar='"'):
user = User.objects.create(username=name)
for submission in submissions.split(','):
title = "".join(submission.strip().split()[:1])
if title in entries_by_title:
entry = entries_by_title[title]
else:
entry = meeting.add_entry(
title=title, duration=datetime.timedelta(minutes=10))
entries_by_title[title] = entry
entry.participations.create(user=user)
#print len(entries_by_title)
setup = 'from ecs.meetings.models import Meeting;m = Meeting.objects.get(pk=%s);f = m.create_evaluation_func(lambda x: x);perm=list(m)' % meeting.pk
print("%.3fms per metric constructor" %
(timeit.timeit('f(perm)', setup, number=1000)))
setup = 'from ecs.utils.genetic_sort import order_crossover'
print("%.3fms per order_crossover" % (timeit.timeit(
'order_crossover((1,2,3,4,5,6,7,8,9), (9,8,7,6,5,4,3,2,1))',
setup,
number=1000)))
setup = 'from ecs.utils.genetic_sort import swap_mutation'
print("%.3fms per swap_mutuation" % (timeit.timeit(
'swap_mutation((1,2,3,4,5,6,7,8,9))', setup, number=1000)))
setup = 'from ecs.utils.genetic_sort import inversion_mutation'
print("%.3fms per swap_mutuation" % (timeit.timeit(
'inversion_mutation((1,2,3,4,5,6,7,8,9))', setup, number=1000)))
setup = 'from ecs.utils.genetic_sort import displacement_mutation'
print("%.3fms per displacement_mutation" % (timeit.timeit(
'displacement_mutation((1,2,3,4,5,6,7,8,9))', setup, number=1000)))
get_metrics = meeting.create_evaluation_func(lambda x: x)
evalf = meeting.create_evaluation_func(lambda metrics: 1000 * 1000 * (
10.0 / (metrics._waiting_time_total + 1) + 0 /
(metrics._waiting_time_max + 1)))
entries, users = meeting.timetable
sorter = GeneticSorter(entries,
evalf,
population_size=100,
crossover_p=0.3,
mutations={
inversion_mutation: 0.002,
swap_mutation: 0.02,
displacement_mutation: 0.01,
})
print("===" * 25)
try:
for i in range(1000):
result = sorter.run(5)
metrics = get_metrics(result)
if not i % 10:
print("%4s %8s %8s %8s %8s %8s %8s %4s %10s" % (
'gen',
'sum',
'avg',
'min',
'max',
'base',
'value',
'pop',
'secs/gen',
))
print("%4d %8d %8d %8d %8d %8d %8d %4d %10.4f" %
(sorter.generation_count,
int(metrics.waiting_time_total.total_seconds() / 60),
int(metrics.waiting_time_avg.total_seconds() / 60),
int(metrics.waiting_time_min.total_seconds() / 60),
int(metrics.waiting_time_max.total_seconds() / 60),
int(evalf(entries)), int(evalf(result)),
len(sorter.population), sorter.time_per_generation))
except KeyboardInterrupt:
pass
transaction.set_rollback(True)
def handle(self, **options):
submission_data = {}
try:
meeting = Meeting.objects.create(start=datetime.datetime(2010, 4, 13, 10), title="Test")
entries_by_title = {}
for name, submissions, comments in csv.reader(
open(os.path.join(os.path.dirname(__file__), "data", "meeting100413.csv"), "r"),
delimiter=",",
quotechar='"',
):
user = User.objects.create(username=name)
for submission in submissions.split(","):
title = "".join(submission.strip().split()[:1])
if title in entries_by_title:
entry = entries_by_title[title]
else:
entry = meeting.add_entry(title=title, duration=datetime.timedelta(minutes=10))
entries_by_title[title] = entry
entry.add_user(user)
# print len(entries_by_title)
setup = (
"from ecs.meetings.models import Meeting;m = Meeting.objects.get(pk=%s);f = m.create_evaluation_func(lambda x: x);perm=list(m)"
% meeting.pk
)
print "%.3fms per metric constructor" % (timeit.timeit("f(perm)", setup, number=1000))
setup = "from ecs.utils.genetic_sort import order_crossover"
print "%.3fms per order_crossover" % (
timeit.timeit("order_crossover((1,2,3,4,5,6,7,8,9), (9,8,7,6,5,4,3,2,1))", setup, number=1000)
)
setup = "from ecs.utils.genetic_sort import swap_mutation"
print "%.3fms per swap_mutuation" % (
timeit.timeit("swap_mutation((1,2,3,4,5,6,7,8,9))", setup, number=1000)
)
setup = "from ecs.utils.genetic_sort import inversion_mutation"
print "%.3fms per swap_mutuation" % (
timeit.timeit("inversion_mutation((1,2,3,4,5,6,7,8,9))", setup, number=1000)
)
setup = "from ecs.utils.genetic_sort import displacement_mutation"
print "%.3fms per displacement_mutation" % (
timeit.timeit("displacement_mutation((1,2,3,4,5,6,7,8,9))", setup, number=1000)
)
get_metrics = meeting.create_evaluation_func(lambda x: x)
evalf = meeting.create_evaluation_func(
lambda metrics: 1000
* 1000
* (10.0 / (metrics._waiting_time_total + 1) + 0 / (metrics._waiting_time_max + 1))
)
entries, users = meeting.timetable
sorter = GeneticSorter(
entries,
evalf,
population_size=100,
crossover_p=0.3,
mutations={inversion_mutation: 0.002, swap_mutation: 0.02, displacement_mutation: 0.01},
)
print "===" * 25
try:
for i in xrange(1000):
result = sorter.run(5)
metrics = get_metrics(result)
if not i % 10:
print "%4s %8s %8s %8s %8s %8s %8s %4s %10s" % (
"gen",
"sum",
"avg",
"min",
"max",
"base",
"value",
"pop",
"secs/gen",
)
print "%4d %8d %8d %8d %8d %8d %8d %4d %10.4f" % (
sorter.generation_count,
int(timedelta_to_seconds(metrics.waiting_time_total) / 60),
int(timedelta_to_seconds(metrics.waiting_time_avg) / 60),
int(timedelta_to_seconds(metrics.waiting_time_min) / 60),
int(timedelta_to_seconds(metrics.waiting_time_max) / 60),
int(evalf(entries)),
int(evalf(result)),
len(sorter.population),
sorter.time_per_generation,
)
except KeyboardInterrupt:
pass
finally:
transaction.rollback()
