def test_hyperband_api():
import random
params = Space.from_dict({'a': 'uniform(0, 1)'})
hpo = Hyperband(Fidelity(0, 100, 10, 'epochs'), params)
assert not hpo.is_done()
for rung in range(3):
params_set = hpo.suggest()
for i, params in enumerate(params_set):
print(i, params)
try:
hpo.suggest()
raise RuntimeError()
except WaitingForTrials:
pass
except OptimizationIsDone:
pass
for i, params in enumerate(params_set):
v = random.uniform(0, 1)
if i == len(params_set) - 1:
v = 1e-10
hpo.observe(params, v)
print('-------')
assert hpo.is_done()
print(hpo.result())
print(hpo.info())
def run(uri, database, namespace, function, fidelity, space, count, variables,
plot_filename, objective, defaults, save_dir='.', sleep_time=60,
register=True):
if fidelity is None:
fidelity = Fidelity(1, 1, name='epoch').to_dict()
defaults.update(variables)
config = {
'name': 'random_search',
'fidelity': fidelity,
'space': space,
'count': count
}
client = new_client(uri, database)
if not is_registered(client, namespace) and register:
register_hpo(client, namespace, function, config, defaults=defaults)
while not is_hpo_completed(client, namespace):
print_status(client, namespace)
time.sleep(sleep_time)
# get the result of the HPO
print(f'HPO is done')
data = fetch_hpo_valid_curves(client, namespace, list(sorted(variables.keys())))
save_results(namespace, data, save_dir)
plot(space, objective, data, plot_filename, model_seed=1)
def build_robo(model_type, n_init=2, count=5):
params = {'x': 'uniform(-5, 10)', 'y': 'uniform(0, 15)'}
return HPOptimizer('robo',
fidelity=Fidelity(0, 100, 10, 'epoch').to_dict(),
space=params,
model_type=model_type,
count=count,
n_init=n_init)
def test_hyperband_simple_sequential():
import random
params = Space.from_dict({'a': 'uniform(0, 1)'})
hpo = Hyperband(Fidelity(0, 100, 10, 'epochs'), params)
for params in hpo:
hpo.observe(params, result=random.uniform(0, 1))
assert hpo.is_done()
print(hpo.result())
print(hpo.info())
def check():
def add(a, b, **kwargs):
return a + b
hpo = HPOptimizer('hyperband',
fidelity=Fidelity(1, 30, 2),
space={
'a': 'uniform(0, 1)',
'b': 'uniform(0, 1)'
})
while not hpo.is_done():
for args in hpo:
# try a new configuration
result = add(**args)
# forward the result to the optimizer
hpo.observe(args, result)
print(hpo.result())
import pytest
import logging
import os
from olympus.utils.log import set_log_level
set_log_level(logging.DEBUG)
from olympus.hpo.worker import HPOWorkGroup, make_remote_call
from olympus.utils.testing import my_trial
from olympus.hpo.fidelity import Fidelity
from olympus.hpo import HPOptimizer
FIDELITY = Fidelity(1, 30, 10).to_dict()
def run_nomad_hpo(hpo_name,
uri,
launch_server=True,
fidelity=FIDELITY,
group='nomad',
workers=10):
"""Worker are converted to HPO when new trials are needed then killed"""
with HPOWorkGroup(uri,
'olympus',
f'classification-{group}-1',
clean=True,
launch_server=launch_server) as group:
group.launch_workers(workers)
params = {
'a': 'uniform(0, 1)',
import numpy
import pytest
from sspace import Space
from robo.fmin import bayesian_optimization
from olympus.hpo import HPOptimizer
from olympus.hpo.fidelity import Fidelity
from olympus.hpo.robo import RoBO, build_bounds
from olympus.hpo.optimizer import WaitingForTrials
FIDELITY = Fidelity(0, 100, 10, 'epoch')
def branin(x, y, epoch=0, uid=None, other=None, experiment_name=None, client=None):
b = (5.1 / (4.*numpy.pi**2))
c = (5. / numpy.pi)
t = (1. / (8.*numpy.pi))
return 1.*(y-b*x**2+c*x-6.)**2+10.*(1-t)*numpy.cos(x)+10.
def branin_for_original_robo(x):
return branin(x[0], x[1])
def get_robo_results(count):
lower = numpy.array([-5, 0])
upper = numpy.array([10, 15])