def setUp(self):
self.expName = 'test_experiment_unit_tests'
self.exp = ro.Experiment(
self.expName, {
'param1': ro.Uniform(low=0.0, high=100.0, dtype='int'),
'param2': ro.Normal(mean=100, std=10, dtype='int'),
})
def run_ro(init, num_epochs):
param = init
e = ro.Experiment('ro_example', {'alpha': ro.Uniform(low=0.0, high=0.01)})
e.sample_all_params()
for epoch in range(num_epochs):
param = param - e.alpha * dloss(param)
e.add_result(loss(param))
return e
def run_hb(init, num_epochs):
param = init
e = ro.HyperBand('hb_example', {'alpha': ro.Uniform(low=0.0, high=0.01)},
num_iter=num_epochs)
e.sample_all_params()
for epoch in range(num_epochs):
param = param - e.alpha * dloss(param)
if e.stop(loss(param)):
return e
e.add_result(loss(param))
return e
#!/usr/bin/env python
import randopt as ro
from bonn import Bonn
def loss(x, y, z):
return x**2 + y**2 + z**2
if __name__ == '__main__':
e = ro.Experiment('bo_simple', {
'x': ro.Choice([0.0, 1, 2, 3, 4, 5, 6, 7]),
'y': ro.Gaussian(0.0, 3.0),
'z': ro.Uniform(0.0, 1.0),
})
bo = Bonn(e)
e.sample_all_params()
res = loss(e.x, e.y, e.z)
e.add_result(res)
for i in xrange(200):
bo.fit()
bo.sample(e)
res = loss(e.x, e.y, e.z)
print res
e.add_result(res, {
'trial': i
})
def test_raise_with_param_named_result(self):
with self.assertRaises(ValueError):
ro.Experiment(
'invalid experiment', {
'result': ro.Uniform(low=0.0, high=100.0, dtype='int'),
})
#!/usr/bin/env python
import randopt as ro
def loss(w, x, y, z):
return w**2 + x**2 + y**2 + z**2
if __name__ == '__main__':
e = ro.Experiment('multi_params_example', {
'dog': ro.Normal(mean=0.0, std=1.0, dtype='float'),
'cat': ro.Uniform(low=-1.0, high=1.0, dtype='float'),
'dolphin': ro.LognormVariate(mean=0.0, std=1.0, dtype='float'),
'any_name': ro.Choice([0.01, 0.05, 0.1, 0.5, 0.7, 0.9], sampler=ro.Uniform()),
})
# Seeding will make all of your searches reproducible. (Usually not wanted)
e.seed(1234)
# Randomly sampling parameters
for i in range(100):
e.sample_all_params()
res = loss(e.dog, e.cat, e.dolphin, e.any_name)
print('Result: ', res)
# Example of using the second parameter
e.add_result(res, data={
'sup.data': [e.dog, e.cat, e.dolphin, e.any_name]
})
# Save/load the state of the random number generators
e.save_state('./multi_params_state.pk')
#!/usr/bin/env python3
import randopt as ro
from random import random
def loss(x, y):
return [(x**2 + y**2 + random()) / i for i in range(1, 51)]
if __name__ == '__main__':
exp = ro.Experiment('quadratic', params={
'x': ro.Gaussian(),
'y': ro.Uniform(-0.5, 0.5)
})
for _ in range(20):
exp.sample_all_params()
conv = loss(exp.x, exp.y)
exp.add_result(conv[-1], data={
'convergence': conv
})