def minimize(self):
# Compass requires an initial point
if self.x0 is None:
self.x0 = self.domain.l + self.domain._range / 2
res = minimizeCompass(self.f, bounds=self.domain.bounds, x0=self.x0,
niter=self.config.niter,
deltatol=self.config.deltatol,
deltainit=self.config.deltainit,
scaling=self.domain.range,
redfactor=self.config.redfactor,
errorcontrol=False,
funcNinit=30,
alpha=0.05,
paired=False)
x0 = res['x']
res = minimizeCompass(self.f, bounds=self.domain.bounds, x0=x0,
niter=self.config.niter,
deltatol=self.config.deltatol,
deltainit=2*self.config.deltatol,
scaling=self.domain.range,
redfactor=self.config.redfactor,
errorcontrol=True,
funcNinit=30,
alpha=0.05,
paired=False)
def noisy_opt():
bounds = [[params[i][1], params[i][2]] for i in range(len(params))]
x0 = np.array([params[i][3] for i in range(len(params))])
res = minimizeCompass(obj_func_one,
bounds=bounds,
x0=x0,
deltatol=0.1,
paired=False)
print(res)
def optimize(stim_name_list, pin_score_dict):
score_mat = np.concatenate(tuple(
[pin_score_dict[n] for n in stim_name_list]),
axis=0)
optimizer_len = np.shape(score_mat)[0]
bound = [[0, 100] for _ in range(optimizer_len)]
initial_guess = np.array(
[np.random.random_sample() * 100 for _ in range(optimizer_len)])
obj = construct_objective(score_mat)
print("starting min compas")
return minimizeCompass(obj,
bounds=bound,
x0=initial_guess,
deltainit=1000,
deltatol=0.001,
paired=False)
def optimizer_prediction(number_of_layers, material, filename):
'''
Optimizes the prediction of the TBR by changing the enrichment fractions and the TBR values and returns the best enrichment configuration for the highest TBR.
'''
bounds=()
x0=[]
for k in range(number_of_layers):
bounds=bounds+((0,1),)
x0.append(0.5)
bounds = bounds + ((0,2),)
print(bounds)
x0.append(1.2)
print(x0)
result = minimizeCompass(find_prediction, bounds=bounds, x0=x0, deltatol=0.01, paired=True, disp=False)
args=[{'number_of_layers' : number_of_layers,'material' : material, 'filename' : filename}]
print(result)
print(1/result.fun)
def simulate(material, first_wall):
results = []
bounds = ()
x0 = []
with open(
'results/result_noisyopt_' + str(material) + '_' +
str(first_wall) + '.json', 'w') as file_object:
json.dump([], file_object, indent=2)
for number_of_layers in [1, 2, 3, 4]:
with open('results/result_noisyopt_' + str(material) + '_' +
str(first_wall) + '.json') as file_object:
results = json.load(file_object)
bounds = bounds + ((0, 1), )
print(bounds)
x0.append(0.5)
print(x0)
result = minimizeCompass(find_tbr,
bounds=bounds,
x0=x0,
deltatol=0.01,
paired=True,
disp=False)
args = [{'material': material, 'first_wall': first_wall}]
print(result)
print(1 / result.fun)
json_output = {
"max_tbr": 1 / result.fun,
"best_enrichment": list(result.x),
"number_of_layers": len(result.x),
"breeder_material_name": args[0]['material'],
"include_first_wall": args[0]['first_wall']
}
print('json_output', json_output)
results.append(json_output)
with open(
'results/result_noisyopt_' + str(material) + '_' +
str(first_wall) + '.json', 'w') as file_object:
json.dump(results, file_object, indent=2)
def optimize(stim_name_list, subset_list=None, min_bound=0, max_bound=100):
if isinstance(stim_name_list, str):
# Single stim optimization.
stim_name_list = [stim_name_list]
else:
# Multi stim optimization.
# Already a list, nothing to be done to stim_name_list.
pass
score_mat = np.concatenate(tuple(
[pin_score_dict[n] for n in stim_name_list]),
axis=0)
print("SCORE MAT NAN PATCH")
score_mat = np.where(
np.isnan(score_mat),
np.ma.array(score_mat, mask=np.isnan(score_mat)).max(axis=0),
score_mat)
# print("SCORE MAT CLIP")
# score_mat = np.clip(score_mat,0, 100)
# Use only a subset of the data points if provided.
# Otherwise, use all of the data points still.
if subset_list is not None:
score_mat = np.take(score_mat, subset_list, 1)
optimizer_len = np.shape(score_mat)[0]
bound = [[min_bound, max_bound] for _ in range(optimizer_len)]
np.random.seed(0)
initial_guess = np.array(
[np.random.random_sample() * 100 for _ in range(optimizer_len)])
obj = construct_objective(score_mat)
return minimizeCompass(obj,
bounds=bound,
x0=initial_guess,
deltainit=100,
deltatol=0.01,
paired=False), score_mat, obj
def optimize(stim_name_list, pin_score_dict, collapsed_sensitivity_dict):
score_mat = np.concatenate(tuple(
[pin_score_dict[n] for n in stim_name_list]),
axis=0)
sensitivity_mat = np.concatenate(tuple(
[collapsed_sensitivity_dict[n] for n in stim_name_list]),
axis=0)
optimizer_len = np.shape(score_mat)[0]
bound = [[0, 100] for _ in range(optimizer_len)]
initial_guess = np.array(
[np.random.random_sample() * 100 for _ in range(optimizer_len)])
obj = construct_objective(score_mat,
sensitivity_mat,
param_select_vec=[9],
obj_comb_vec=[1, 1])
return minimizeCompass(obj,
bounds=bound,
x0=initial_guess,
deltainit=1000,
deltatol=1,
paired=False)
ubMatrix = UpperBounds
bRangeMatrix = ubMatrix - lbMatrix
x0 = lbMatrix + np.random.rand(1, nVars) * bRangeMatrix
# print(x0[0])
# obj_sim = Simulator(ObjectiveFcn(x0[0],nVars,currentStock,currentOrder))
# minimizeCompass(obj_sim.simulate, [0, 0], method='BFGS', callback=obj_sim.callback, options={"disp": True})
res = None
while res is None:
try:
# connect
res = minimizeCompass(
ObjectiveFcn,
x0=x0[0],
deltatol=0.1,
paired=False,
args=[nVars, currentStock, currentOrder],
errorcontrol=False)
except:
print(
"\nAn unexcpected error occured. Program will terminate.\nPlease try running again..."
)
time.sleep(2)
sys.exit()
# print(f"Number of calls to Simulator instance {obj_sim.num_calls}")
pos = res.x
iterationsList.append(res.nit)
logfile_name = start_time.strftime('test_noisyopt_results' +
'%Y%m%d_%H%M%S')
print("Log file: %s" % logfile_name)
logfile = open(logfile_name, "w")
log(logfile, 'Optimization method: noisyopt', stdout=False)
log(logfile, "0-dB crossing\trmse\t\tExec Time")
if args.func == 'quad_func':
bnds = [(-1, 1)]
x0 = [1.0]
save = True
res = minimizeCompass(quad_func,
x0,
args=(save, save),
bounds=bnds,
deltainit=deltainit,
funcNinit=funcNinit,
paired=False,
feps=feps,
disp=False)
x = np.linspace(-1, 1, 100)
for i in x:
y.append(quad_func(i, False))
elif args.func == 'cavity_step_rmse':
bnds = [(10000, 70000)]
x0 = [10000]
res = minimizeCompass(cavity_step_rmse,
x0,
args=(args.time, args.time),
# This is a minimal usage example for the optimization routines
# see also http://noisyopt.readthedocs.io/en/latest/tutorial.html
import numpy as np
from noisyopt import minimizeCompass, minimizeSPSA
# definition of noisy function to be optimized
def obj(x):
return (x**2).sum() + 0.1 * np.random.randn()
bounds = [[-3.0, 3.0], [0.5, 5.0]]
x0 = np.array([-2.0, 2.0])
res = minimizeCompass(obj, bounds=bounds, x0=x0, deltatol=0.1, paired=False)
print(res)
res = minimizeSPSA(obj, bounds=bounds, x0=x0, niter=1000, paired=False)
print(res)
print(Y)
#readLog() ##We are doing nothing here for now
##converting polar displacemt to polarization for computing error
##Not required for now ; comparing the polar displacements with DFT polar displacements
#Y[6]=Y[6]*CbyV/((Y[0]+1)*(Y[1]+1)*(Y[2]+1))
#Y[7]=Y[7]*CbyV/((Y[0]+1)*(Y[1]+1)*(Y[2]+1))
#Y[8]=Y[8]*CbyV/((Y[0]+1)*(Y[1]+1)*(Y[2]+1))
##script to compute error
totalError = 0
for i in [0, 1, 2, 8]:
totalError = totalError + ((Y[i] - Y0[i]) / (Y0[i]))**2
IT = IT + 1
return (totalError)
#paramWrite(pInitFName,p0) ##note required, defined n the function itself
#fopt.Bounds(lB, uB, keep_feasible=False)
#fopt = optimize.minimize(Fun,p0,method='L-BFGS-B')
fopt = minimizeCompass(Fun, bounds=pBounds, x0=p0, deltatol=0.1, paired=False)
#fopt.Bounds(lB, uB, keep_feasible=False)
#printing
print("###################################################################")
print("Minimum function value =", fopt.fun)
print("Minimum function value at x =", fopt.x)
print("Number of iterations done =", IT)
#print(xopt)
print("###################################################################")
# In[ ]:
def test_minimizeCompass():
deltatol = 1e-3
## basic testing without stochasticity
def quadratic(x):
return (x**2).sum()
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.5, 1.0]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
# test with callback
def callback(x):
assert len(x) == 2
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.5, 1.0]),
deltatol=deltatol,
errorcontrol=False,
callback=callback)
# test with scaling
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.5, 1.0]),
deltatol=deltatol,
scaling=np.array([0.1, 1.0]),
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
# test with output
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.5, 1.0]),
deltatol=deltatol,
errorcontrol=False,
disp=True)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic,
np.asarray([2.5, -3.2]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic,
np.asarray([2.5, -3.2, 0.9, 10.0, -0.3]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, np.zeros(5), atol=deltatol)
npt.assert_equal(res.free, [False, False, False, False, False])
## test bound handling
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.5, 0.5]),
bounds=np.asarray([[0, 1], [0, 1]]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic,
np.asarray([0.8, 0.8]),
bounds=np.asarray([[0.5, 1], [0.5, 1]]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.5, 0.5], atol=deltatol)
npt.assert_equal(res.free, [False, False])
## test args passing
def quadratic_factor(x, factor):
return factor * (x**2).sum()
res = noisyopt.minimizeCompass(quadratic_factor,
np.asarray([0.5, 1.0]),
paired=False,
args=(1.0, ))
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
## test determination of unconstrained variables
def quadratic_except_last(x):
return (x[:-1]**2).sum()
res = noisyopt.minimizeCompass(quadratic_except_last,
np.asarray([0.5, 1.0]),
errorcontrol=False)
npt.assert_approx_equal(res.x[0], 0.0)
npt.assert_equal(res.free, [False, True])
## test errorcontrol for stochastic function
def stochastic_quadratic(x, seed=None):
prng = np.random if seed is None else np.random.RandomState(seed)
return (x**2).sum() + prng.randn(1) + 0.5 * np.random.randn(1)
deltatol = 0.5
# test unpaired
res = noisyopt.minimizeCompass(stochastic_quadratic,
np.array([4.55, 3.0]),
deltainit=2.0,
deltatol=deltatol,
errorcontrol=True,
paired=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
# test paired
res = noisyopt.minimizeCompass(stochastic_quadratic,
np.array([4.55, 3.0]),
deltainit=2.0,
deltatol=deltatol,
errorcontrol=True,
paired=True)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
### noisyopt start
objcount = 0
def obj(x):
global objcount
objcount += 1
#if objcount%1000==0:
# print(objcount)
a = x[0]
b = x[1]
c = x[2]
d = x[3]
lib.c_assignAndRun(a, b, c, d)
return getCurEnergy()
#return getCurMoment()
#start_time = time.time()
#print(obj([12,6]))
#print("--- %s seconds ---" % (time.time() - start_time))
lib.c_setNumConfigs(100)
res = minimizeCompass(obj, x0=snx0, deltatol=1, paired=False)
print(objcount)
print(res.success)
print(snx0)
print(res.x)
np.savetxt(resfile, np.vstack((snx0, res.x)))
xx = np.eye(2**n_qubits)
def Los2(Theta):
return sum(
np.linalg.norm(func(Theta, xx[i]) - xx[i])
for i in range(2**n_qubits - 1))
def Los3(test):
return Los2([0, 0, test, 0, 0, 0])
res = minimizeCompass(Los3,
x0=[1],
bounds=[[-np.pi, np.pi]],
deltatol=0.1,
paired=False)
a = {
"00": func([0, 0, 0, 0, 0, 0], [1, 0, 0, 0])[0],
"01": func([0, 0, 0, 0, 0, 0], [0, 1, 0, 0])[1],
"10": func([0, 0, 0, 0, 0, 0], [0, 0, 1, 0])[2],
"11": func([0, 0, 0, 0, 0, 0], [0, 0, 0, 1])[3],
}
b = {
"00": func([0, 0, -0.5, 0, 0, 0], [1, 0, 0, 0])[0],
"01": func([0, 0, -0.5, 0, 0, 0], [0, 1, 0, 0])[1],
"10": func([0, 0, -0.5, 0, 0, 0], [0, 0, 1, 0])[2],
"11": func([0, 0, -0.5, 0, 0, 0], [0, 0, 0, 1])[3],
}
def test_minimizeCompass():
deltatol = 1e-3
## basic testing without stochasticity
def quadratic(x):
return (x**2).sum()
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.5, 1.0]), deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
# test with callback
def callback(x):
assert len(x) == 2
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.5, 1.0]), deltatol=deltatol,
errorcontrol=False, callback=callback)
# test with scaling
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.5, 1.0]), deltatol=deltatol,
scaling=np.array([0.1, 1.0]),
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
# test with output
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.5, 1.0]), deltatol=deltatol,
errorcontrol=False, disp=True)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic, np.asarray([2.5, -3.2]), deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic, np.asarray([2.5, -3.2, 0.9, 10.0, -0.3]),
deltatol=deltatol, errorcontrol=False)
npt.assert_allclose(res.x, np.zeros(5), atol=deltatol)
npt.assert_equal(res.free, [False, False, False, False, False])
## test bound handling
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.5, 0.5]),
bounds=np.asarray([[0, 1], [0, 1]]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
res = noisyopt.minimizeCompass(quadratic, np.asarray([0.8, 0.8]),
bounds=np.asarray([[0.5, 1], [0.5, 1]]),
deltatol=deltatol,
errorcontrol=False)
npt.assert_allclose(res.x, [0.5, 0.5], atol=deltatol)
npt.assert_equal(res.free, [False, False])
## test args passing
def quadratic_factor(x, factor):
return factor*(x**2).sum()
res = noisyopt.minimizeCompass(quadratic_factor, np.asarray([0.5, 1.0]),
paired=False, args=(1.0,))
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
## test determination of unconstrained variables
def quadratic_except_last(x):
return (x[:-1]**2).sum()
res = noisyopt.minimizeCompass(quadratic_except_last, np.asarray([0.5, 1.0]),
errorcontrol=False)
npt.assert_approx_equal(res.x[0], 0.0)
npt.assert_equal(res.free, [False, True])
## test errorcontrol for stochastic function
def stochastic_quadratic(x, seed=None):
prng = np.random if seed is None else np.random.RandomState(seed)
return (x**2).sum() + prng.randn(1) + 0.5*np.random.randn(1)
deltatol = 0.5
# test unpaired
res = noisyopt.minimizeCompass(stochastic_quadratic, np.array([4.55, 3.0]),
deltainit=2.0, deltatol=deltatol,
errorcontrol=True, paired=False)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
# test paired
res = noisyopt.minimizeCompass(stochastic_quadratic, np.array([4.55, 3.0]),
deltainit=2.0, deltatol=deltatol,
errorcontrol=True, paired=True)
npt.assert_allclose(res.x, [0.0, 0.0], atol=deltatol)
npt.assert_equal(res.free, [False, False])
# This is a minimal usage example for the optimization routines
# see also http://noisyopt.readthedocs.io/en/latest/tutorial.html
import numpy as np
from noisyopt import minimizeCompass, minimizeSPSA
# definition of noisy function to be optimized
def obj(x):
return (x**2).sum() + 0.1*np.random.randn()
bounds = [[-3.0, 3.0], [0.5, 5.0]]
x0 = np.array([-2.0, 2.0])
res = minimizeCompass(obj, bounds=bounds, x0=x0, deltatol=0.1, paired=False)
print(res)
res = minimizeSPSA(obj, bounds=bounds, x0=x0, niter=1000, paired=False)
print(res)
