def __round_optim(self, obj, x0, cns):
"""
Solves the iteration round's local optimization
:param obj: the approximate objective
:param x0: initial point of the optimization, i.e., the round's sampled point
:param cns: constraints
:return: result of the optimization
"""
res = None
if self.optimizer == "scipy":
from scipy.optimize import minimize
if (len(cns) > 0) and (self.scipy_solver
not in ["COBYLA", "SLSQP", "L-BFGS-B"]):
raise ValueError("%s does not support arbitrary constraints" %
self.scipy_solver)
res = minimize(obj,
x0,
method=self.scipy_solver,
constraints=cns,
bounds=self.bounds)
else:
try:
from Optimithon import Base, QuasiNewton
from numpy import inf
cns_ = [_["fun"] for _ in cns]
if self.bounds is not None:
for i in range(len(self.bounds)):
if self.bounds[i][0] != -inf:
cns_.append(
lambda x, i_=i: x[i_] - self.bounds[i_][0])
if self.bounds[i][1] != inf:
cns_.append(
lambda x, i_=i: self.bounds[i_][1] - x[i_])
optm = Base(
obj,
ineq=cns_,
x0=x0,
br_func=self.optimithon_br_func,
penalty=self.optimithon_penalty,
method=QuasiNewton,
t_method=self.optimithon_t_method,
dd_method=self.optimithon_dd_method,
ls_method=self.optimithon_ls_method,
ls_bt_method=self.optimithon_ls_bt_method,
max_iter=self.optimithon_max_iter,
difftool=self.optimithon_difftool,
)
optm.Verbose = False
optm()
res = optm.solution
res.fun = res.objective
except ModuleNotFoundError:
pass
return res
from scipy.optimize import minimize
NumVars = 9
fun = lambda x: sum([100 * (x[i + 1] - x[i]**2)**2 +
(1 - x[i])**2 for i in range(NumVars - 1)])
cons = [
{'type': 'ineq', 'fun': lambda x: 9 - x[i]**2} for i in range(NumVars)]
x0 = array([0 for _ in range(NumVars)])
sol1 = minimize(fun, x0, method='COBYLA', constraints=cons)
sol2 = minimize(fun, x0, method='SLSQP', constraints=cons)
print("solution according to 'COBYLA':")
print(sol1)
print("solution according to 'SLSQP':")
print(sol2)
OPTIM = Base(fun, ineq=[lambda x: 9 - x[i]**2 for i in range(NumVars)],
br_func='Carrol',
penalty=1.e6,
method=QuasiNewton, x0=x0, # max_lngth=100.,
t_method='Cauchy_x', # 'Cauchy_x', 'ZeroGradient',
dd_method='BFGS',
# 'Newton', 'SR1', 'HestenesStiefel', 'PolakRibiere', 'FletcherReeves', 'Gradient', 'DFP', 'BFGS', 'Broyden', 'DaiYuan'
ls_method='Backtrack', # 'BarzilaiBorwein', 'Backtrack',
ls_bt_method='Armijo', # 'Armijo', 'Goldstein', 'Wolfe', 'BinarySearch'
)
OPTIM.Verbose = False
OPTIM.MaxIteration = 1500
OPTIM()
print("================================================="*2)
print(OPTIM.solution)
from Optimithon import Base
from Optimithon import QuasiNewton
from numpy import array, sin, pi
from scipy.optimize import minimize
fun = lambda x: sin(x[0] + x[1]) + (x[0] - x[1]) ** 2 - 1.5 * x[0] + 2.5 * x[1] + 1.
x0 = array((0., 0.))
print(fun(x0))
sol1 = minimize(fun, x0, method='COBYLA')
sol2 = minimize(fun, x0, method='SLSQP')
print("solution according to 'COBYLA':")
print(sol1)
print("solution according to 'SLSQP':")
print(sol2)
OPTIM = Base(fun,
method=QuasiNewton, x0=x0, # max_lngth=100.,
t_method='Cauchy_x', # 'Cauchy_x', 'ZeroGradient',
dd_method='BFGS',
# 'Newton', 'SR1', 'HestenesStiefel', 'PolakRibiere', 'FletcherReeves', 'Gradient', 'DFP', 'BFGS', 'Broyden', 'DaiYuan'
ls_method='Backtrack', # 'BarzilaiBorwein', 'Backtrack',
ls_bt_method='Armijo', # 'Armijo', 'Goldstein', 'Wolfe', 'BinarySearch'
)
OPTIM.Verbose = False
OPTIM.MaxIteration = 1500
OPTIM()
print("==========================" * 4)
print(OPTIM.solution)