def test_bandit_json_3(self):
self.cmd = ("bandit_json evaluate", "hyperopt.bandits.gauss_wave2")
self.exp_keys = ["key0"]
self.bandit = gauss_wave2()
self.use_stop = True
self.n_threads = 3
self.jobs_per_thread = 2
self.work()
def test_bandit_json_3(self):
self.cmd = ('bandit_json evaluate', 'hyperopt.bandits.gauss_wave2')
self.exp_keys = ['key0']
self.bandit = gauss_wave2()
self.use_stop = True
self.n_threads = 3
self.jobs_per_thread = 2
self.work()
def test_bandit_json_3(self):
self.cmd = ('bandit_json evaluate',
'hyperopt.bandits.gauss_wave2')
self.exp_keys = ['key0']
self.bandit = gauss_wave2()
self.use_stop = True
self.n_threads = 3
self.jobs_per_thread = 2
self.work()
def test_driver_attachment_1(self):
bandit_name = 'hyperopt.bandits.gauss_wave2'
bandit_args = ()
bandit_kwargs = {}
blob = cPickle.dumps((bandit_name, bandit_args, bandit_kwargs))
def prep_trials(trials):
print 'storing attachment'
trials.attachments['aname'] = blob
assert trials.attachments['aname'] == blob
assert trials.attachments[u'aname'] == blob
self.prep_trials = prep_trials
self.cmd = ('driver_attachment', 'aname')
self.exp_keys = ['key0']
self.bandit = gauss_wave2()
self.use_stop = False
self.n_threads = 3
self.jobs_per_thread = 2
self.work()
def test_driver_attachment_1(self):
bandit_name = "hyperopt.bandits.gauss_wave2"
bandit_args = ()
bandit_kwargs = {}
blob = cPickle.dumps((bandit_name, bandit_args, bandit_kwargs))
def prep_trials(trials):
print "storing attachment"
trials.attachments["aname"] = blob
assert trials.attachments["aname"] == blob
assert trials.attachments[u"aname"] == blob
self.prep_trials = prep_trials
self.cmd = ("driver_attachment", "aname")
self.exp_keys = ["key0"]
self.bandit = gauss_wave2()
self.use_stop = False
self.n_threads = 3
self.jobs_per_thread = 2
self.work()
def test_seeds_AAB(self):
# launch 3 simultaneous experiments with seeds A, A, B.
# Verify all experiments run to completion.
# Verify first two experiments run identically.
# Verify third experiment runs differently.
exp_keys = ['A0', 'A1', 'B']
seeds = [1, 1, 2]
n_workers = 2
jobs_per_thread = 6
# -- total jobs = 2 * 6 = 12
# -- divided by 3 experiments: 4 jobs per fmin
max_evals = (n_workers * jobs_per_thread) // len(exp_keys)
# -- should not matter which bandit is used here
bandit = gauss_wave2()
cPickle.dumps(bandit.expr)
cPickle.dumps(passthrough)
worker_threads = [
threading.Thread(
target=TestExperimentWithThreads.worker_thread_fn,
args=(('hostname', ii), jobs_per_thread, 30.0))
for ii in range(n_workers)]
with TempMongo() as tm:
mj = tm.mongo_jobs('foodb')
trials_list = [
MongoTrials(tm.connection_string('foodb'), key)
for key in exp_keys]
fmin_threads = [
threading.Thread(
target=TestExperimentWithThreads.fmin_thread_fn,
args=(bandit.expr, trials, max_evals, seed))
for seed, trials in zip(seeds, trials_list)]
try:
[th.start() for th in worker_threads + fmin_threads]
finally:
print('joining worker threads...')
[th.join() for th in worker_threads + fmin_threads]
# -- not using an exp_key gives a handle to all the trials
# in foodb
all_trials = MongoTrials(tm.connection_string('foodb'))
self.assertEqual(len(all_trials), n_workers * jobs_per_thread)
# Verify that the fmin calls terminated correctly:
for trials in trials_list:
self.assertEqual(
trials.count_by_state_synced(JOB_STATE_DONE),
max_evals)
self.assertEqual(
trials.count_by_state_unsynced(JOB_STATE_DONE),
max_evals)
self.assertEqual(len(trials), max_evals)
# Verify that the first two experiments match.
# (Do these need sorting by trial id?)
trials_A0, trials_A1, trials_B0 = trials_list
self.assertEqual(
[t['misc']['vals'] for t in trials_A0.trials],
[t['misc']['vals'] for t in trials_A1.trials])
# Verify that the last experiment does not match.
# (Do these need sorting by trial id?)
self.assertNotEqual(
[t['misc']['vals'] for t in trials_A0.trials],
[t['misc']['vals'] for t in trials_B0.trials])
import unittest
from hyperopt.base import Trials, trials_from_docs, miscs_to_idxs_vals
from hyperopt import rand
from hyperopt.tests.test_base import Suggest_API
from hyperopt.bandits import gauss_wave2, coin_flip
TestRand = Suggest_API.make_tst_class(rand.suggest, gauss_wave2(), 'TestRand')
class TestRand(unittest.TestCase):
def test_seeding(self):
# -- assert that the seeding works a particular way
domain = coin_flip()
docs = rand.suggest(range(10), domain, Trials(), seed=123)
trials = trials_from_docs(docs)
idxs, vals = miscs_to_idxs_vals(trials.miscs)
# Passes Nov 8 / 2013
self.assertEqual(list(idxs['flip']), range(10))
self.assertEqual(list(vals['flip']), [0, 1, 0, 0, 0, 0, 0, 1, 1, 0])
# -- TODO: put in a test that guarantees that
# stochastic nodes are sampled in a paricular order.
def test_seeds_AAB(self):
# launch 3 simultaneous experiments with seeds A, A, B.
# Verify all experiments run to completion.
# Verify first two experiments run identically.
# Verify third experiment runs differently.
exp_keys = ['A0', 'A1', 'B']
seeds = [1, 1, 2]
n_workers = 2
jobs_per_thread = 6
# -- total jobs = 2 * 6 = 12
# -- divided by 3 experiments: 4 jobs per fmin
max_evals = (n_workers * jobs_per_thread) // len(exp_keys)
# -- should not matter which bandit is used here
bandit = gauss_wave2()
cPickle.dumps(bandit.expr)
cPickle.dumps(passthrough)
worker_threads = [
threading.Thread(target=TestExperimentWithThreads.worker_thread_fn,
args=(('hostname', ii), jobs_per_thread, 30.0))
for ii in range(n_workers)
]
with TempMongo() as tm:
mj = tm.mongo_jobs('foodb')
trials_list = [
MongoTrials(tm.connection_string('foodb'), key)
for key in exp_keys
]
fmin_threads = [
threading.Thread(
target=TestExperimentWithThreads.fmin_thread_fn,
args=(bandit.expr, trials, max_evals, seed))
for seed, trials in zip(seeds, trials_list)
]
try:
[th.start() for th in worker_threads + fmin_threads]
finally:
print('joining worker threads...')
[th.join() for th in worker_threads + fmin_threads]
# -- not using an exp_key gives a handle to all the trials
# in foodb
all_trials = MongoTrials(tm.connection_string('foodb'))
self.assertEqual(len(all_trials), n_workers * jobs_per_thread)
# Verify that the fmin calls terminated correctly:
for trials in trials_list:
self.assertEqual(trials.count_by_state_synced(JOB_STATE_DONE),
max_evals)
self.assertEqual(
trials.count_by_state_unsynced(JOB_STATE_DONE), max_evals)
self.assertEqual(len(trials), max_evals)
# Verify that the first two experiments match.
# (Do these need sorting by trial id?)
trials_A0, trials_A1, trials_B0 = trials_list
self.assertEqual([t['misc']['vals'] for t in trials_A0.trials],
[t['misc']['vals'] for t in trials_A1.trials])
# Verify that the last experiment does not match.
# (Do these need sorting by trial id?)
self.assertNotEqual([t['misc']['vals'] for t in trials_A0.trials],
[t['misc']['vals'] for t in trials_B0.trials])
def test_gausswave2(self):
self.bandit = gauss_wave2()
self.work()
def test_gausswave2(self):
self.bandit = gauss_wave2()
self.work()
import unittest
from hyperopt.base import Trials, trials_from_docs, miscs_to_idxs_vals
from hyperopt import rand
from hyperopt.tests.test_base import Suggest_API
from hyperopt.bandits import gauss_wave2, coin_flip
TestRand = Suggest_API.make_tst_class(rand.suggest, gauss_wave2(), 'TestRand')
class TestRand(unittest.TestCase):
def test_seeding(self):
# -- assert that the seeding works a particular way
domain = coin_flip()
docs = rand.suggest(range(10), domain, Trials(), seed=123)
trials = trials_from_docs(docs)
idxs, vals = miscs_to_idxs_vals(trials.miscs)
# Passes Nov 8 / 2013
self.assertEqual(list(idxs['flip']), range(10))
self.assertEqual(list(vals['flip']), [0, 1, 0, 0, 0, 0, 0, 1, 1, 0])
# -- TODO: put in a test that guarantees that
# stochastic nodes are sampled in a paricular order.
