def main():
bb.search(f=fun, # given function
box=[[0., 1.], [0., 1.], [0., 1.], [0., 1.], [0., 1.], [0., 1.]], # range of values for each parameter
n=40, # number of function calls on initial stage (global search)
m=40, # number of function calls on subsequent stage (local search)
batch=1, # number of calls that will be evaluated in parallel
resfile='output.csv') # text file where results will be saved
def main():
start_time = time.time()
thresholdWavgDistanceToMidpointPercent60minRange = [-1, 1]
thresholdWavgDistanceToMidpointPercent5minRange = [-1, 1]
thresholdWavgDistanceToMidpointPercent30minRange = [-1, 1]
thresholdPowerImbalanceRange = [-1, 3]
thresholdEstimatedFutureWavg5Range = [-1, 2]
thresholdEstimatedFutureWavg30Range = [-1, 2]
thresholdEstimatedFutureWavg60Range = [-1, 2]
thresholdEstimatedFutureWavg120Range = [-1, 2]
thresholdSumFeaturesRange = [5, 6]
thresholdBuyProfitRange = [0.001, 0.0065]
bb.search(
f=blackbox_gekko, # given function
box=[
thresholdWavgDistanceToMidpointPercent60minRange,
thresholdWavgDistanceToMidpointPercent5minRange,
thresholdWavgDistanceToMidpointPercent30minRange,
thresholdPowerImbalanceRange, thresholdEstimatedFutureWavg5Range,
thresholdEstimatedFutureWavg30Range,
thresholdEstimatedFutureWavg60Range,
thresholdEstimatedFutureWavg120Range, thresholdSumFeaturesRange,
thresholdBuyProfitRange
], # range of values for each parameter
n=50, # number of function calls on initial stage (global search)
m=50, # number of function calls on subsequent stage (local search)
batch=8, # number of calls that will be evaluated in parallel
resfile=gekko_base_path + 'blackbox/' + gekko_strategy +
'/output.csv') # text file where results will be saved
print("--- %s seconds ---" % (time.time() - start_time))
def main():
bb.search(
f=predicted_revenue, # given function
box=[[0., 10.], [0., 12.], [0., 40.], [0., 50.],
[0., 100.]], # range of values for each parameter (2D case)
n=100, # number of function calls on initial stage (global search)
m=100, # number of function calls on subsequent stage (local search)
batch=16, # number of calls that will be evaluated in parallel
resfile='revenue_output.csv') # text file where results will be saved
def main():
start_time = time.time()
thresholdSumFeaturesForBuy = [3, 8]
thresholdSumFeaturesForSell = [-3, 5]
bb.search(
f=blackbox_gekko, # given function
box=[thresholdSumFeaturesForBuy, thresholdSumFeaturesForSell
], # range of values for each parameter
n=100, # number of function calls on initial stage (global search)
m=100, # number of function calls on subsequent stage (local search)
batch=12, # number of calls that will be evaluated in parallel
resfile=gekko_base_path + 'blackbox/' + gekko_strategy +
'/output.csv') # text file where results will be saved
print("--- %s seconds ---" % (time.time() - start_time))
def start_blackbox_optimise(minX, maxX):
bb.search(
f=func, # given function
box=[[minX, maxX]], # range of values for each parameter (2D case)
n=10, # number of function calls on initial stage (global search)
m=40, # number of function calls on subsequent stage (local search)
batch=2, # number of calls that will be evaluated in parallel
resfile='output.csv') # text file where results will be saved
print("@@ Done blackbox search")
# End off by sending result
with open('output.csv', 'r') as f:
# line zero is the header, line one is the best result
line1 = f.readlines()[1]
x1, y1 = eval(line1)
# remember to invert back the y value
optimised = -y1
func_queue.put((optimised, False))
def test_2():
"""This function tests that the BLACKBOX results are unaffected by the size of the batch."""
d, box, n, m, batch, strategy = get_valid_request()
rslt_base = None
for batch in range(1, 5):
rslt = search(rosen, box, n, m, batch, strategy)
if rslt_base is None:
rslt_base = rslt
np.testing.assert_almost_equal(rslt, rslt_base)
def test_6():
"""This test function runs a subset of the regression tests."""
vault = pkl.load(
open(TEST_RESOURCES + '/regression_vault.blackbox.pkl', 'rb'))
for idx in np.random.choice(len(vault), size=5):
request, rslt = vault[idx]
# We need to adjust the strategy if there is no MPI available.
if request[-1] not in EXECUTORS:
request[-1] = 'mp'
stat = search(*request)
np.testing.assert_equal(rslt, stat)
def test_1():
"""This function tests that the BLACKBOX results are unaffected by the parallelization
strategy."""
d = np.random.randint(2, 5)
box = [[-10., 10.]] * d
n = d + np.random.randint(5, 10)
batch = np.random.randint(1, 5)
m = np.random.randint(5, 25)
rslt_base = None
for strategy in ['mpi', 'mp']:
rslt = search(rosen, box, n, m, batch, strategy)
if rslt_base is None:
rslt_base = rslt
np.testing.assert_almost_equal(rslt, rslt_base)
def test_3():
"""This function tests that the results line up with the original algorithm."""
# Unfortunately, there will be small differences when using the FORTRAN implementation
# of selected functions. However, this test always passes when the PYTHON versions are
# used. I test all implementations against each other in other tests and as the interface
# is identical, I am confident that the discrepancies here are only due to small numerical
# differences in the PYTHON and FORTRAN calculations that accumulate if all are used at once.
open(PYTHON_FNAME, 'a').close()
d, box, n, m, batch, strategy = get_valid_request()
np.random.seed(123)
alg_original = bb_search(rosen, box, n, m, batch, resfile='output.csv')
alg_revised = search(rosen, box, n, m, batch, strategy, legacy=True)
np.testing.assert_almost_equal(alg_original, alg_revised)
os.remove(PYTHON_FNAME)
def test_5():
"""This test function ensures that the results are unaffected by using either the FORTRAN or
PYTHON function."""
d, box, n, m, batch, strategy = get_valid_request()
rslt_base = None
for is_python in [True, False]:
if is_python:
open(PYTHON_FNAME, 'a').close()
rslt = search(rosen, box, n, m, batch, strategy)
if rslt_base is None:
rslt_base = rslt
np.testing.assert_almost_equal(rslt, rslt_base)
if os.path.exists(PYTHON_FNAME):
os.remove(PYTHON_FNAME)
# grid search
"""
forest_params = {'max_depth': DEPTH_RANGE,'max_features': range(1,feats)}
forest_grid = GridSearchCV(forest, forest_params, cv=5, n_jobs=-1, verbose=True, scoring='f1')
forest_grid.fit(X_train, y_train)
# forest_grid.best_params_
best_scores_grid_search.append(forest_grid.best_score_ )
"""
# blackbox optimization
print("START")
bb.search(
f=bb_func, # given function
box=[DEPTH_RANGE,
range(1, feats)], # range of values for each parameter (2D case)
n=4, # number of function calls on initial stage (global search)
m=2, # number of function calls on subsequent stage (local search)
batch=1, # number of calls that will be evaluated in parallel
resfile=GRAPH_PATH +
'output.csv') # text file where results will be saved
print("...")
print(best_scores_grid_search)
print(sample_range)
plt.plot(sample_range, best_scores_grid_search, label='Grid search')
plt.title('Optimization scores')
plt.savefig(GRAPH_PATH + 'forest_optimization_scores.png', dpi=900)
plt.legend()
from blackbox import search from scipy.optimize import rosen import cProfile import pstats d = 2 box = [[-10, 10]] * d n = 100 m = 10 strategy = 'mp' batch = 2 search(rosen, box, n, m, batch, strategy)
C0 = 299792458
f_0 = 6.4896e9
lambda0 = C0 / f_0
bounds = [
[6.0429e-03 / 1.15, 6.0429e-03 * 1.15], # radius
[1.0871e-02 / 1.15, 1.0871e-02 * 1.15], #length
[4.4621e-03 / 1.15, 4.4621e-03 * 1.15], #pitch2
[4.2901e-03 / 1.5, 4.2901e-03 * 1.5] #pitch1
]
class Mapper:
def map(self, *args, **kwargs):
global client
return client.gather(client.map(*args, **kwargs))
mapper = Mapper()
@contextmanager
def executor():
yield mapper
bb.search(
f=fun, # given function
box=bounds, # range of values for each parameter
n=80, # number of function calls on initial stage (global search)
m=24, # number of function calls on subsequent stage (local search)
batch=8, # number of calls that will be evaluated in parallel
resfile='output.csv', # text file where results will be saved
executor=executor)
import os
from scipy.optimize import rosen
import numpy as np
from blackbox.tests.auxiliary import get_valid_request
from blackbox import search
import blackbox
is_creation = False
if is_creation:
num_tests = 100
tests = []
for seed in range(num_tests):
request = [rosen] + list(get_valid_request())[1:]
rslt = search(*request)
tests.append([request, rslt])
pkl.dump(tests, open('regression_vault.blackbox.pkl', 'wb'))
# TODO: These are only working on heracles at the moment, portability is still an issue.
FNAME_VAULT = os.path.dirname(
blackbox.__file__) + '/tests/material/regression_vault.blackbox.pkl'
for i, test in enumerate(pkl.load(open(FNAME_VAULT, 'rb'))):
print(" running test " + str(i))
request, rslt = test
stat = search(*request)
np.testing.assert_equal(rslt, stat)
