def __init__(self, func, xpar, xmin, xmax, npop, sfactor, xprob, step,
seed):
Opt.__init__(self, func, xmin, xmax)
self.ncores_nm = ncoresNelderMead()
self.key2 = Key2()
self.npop = min(npop, 4096)
self.seed = seed
self.strategies = \
(Strategy0(self.func, self.npar, npop, sfactor, xprob),
Strategy1(self.func, self.npar, npop, sfactor, xprob),
Strategy2(self.func, self.npar, npop, sfactor, xprob),
Strategy3(self.func, self.npar, npop, sfactor, xprob),
Strategy4(self.func, self.npar, npop, sfactor, xprob),
Strategy5(self.func, self.npar, npop, sfactor, xprob),
Strategy6(self.func, self.npar, npop, sfactor, xprob),
Strategy7(self.func, self.npar, npop, sfactor, xprob),
Strategy8(self.func, self.npar, npop, sfactor, xprob),
Strategy9(self.func, self.npar, npop, sfactor, xprob))
xpar = numpy.asarray(xpar)
if step is None:
step = xpar * 1.2 + 1.2
factor = 10
self.polytope = \
SimplexRandom(func, npop, xpar, xmin, xmax, step, seed, factor)
self.local_opt = self.ncores_nm.algo
return
def __call__(self, x, maxnfev, tol, step, finalsimplex, verbose):
npar = len(x)
factor = 2
simplex = SimplexRandom(self.func, npar + 1, x, self.xmin,
self.xmax, None, None, factor)
result = \
self.optimize(x, simplex, maxnfev, tol, finalsimplex, verbose)
# print('MyNelderMead7::__call__ result =', result)
return result
class MyDifEvo(Opt):
def __init__(self, func, xpar, xmin, xmax, npop, sfactor, xprob, step,
seed):
Opt.__init__(self, func, xmin, xmax)
self.ncores_nm = ncoresNelderMead()
self.key2 = Key2()
self.npop = min(npop, 4096)
self.seed = seed
self.strategies = \
(Strategy0(self.func, self.npar, npop, sfactor, xprob),
Strategy1(self.func, self.npar, npop, sfactor, xprob),
Strategy2(self.func, self.npar, npop, sfactor, xprob),
Strategy3(self.func, self.npar, npop, sfactor, xprob),
Strategy4(self.func, self.npar, npop, sfactor, xprob),
Strategy5(self.func, self.npar, npop, sfactor, xprob),
Strategy6(self.func, self.npar, npop, sfactor, xprob),
Strategy7(self.func, self.npar, npop, sfactor, xprob),
Strategy8(self.func, self.npar, npop, sfactor, xprob),
Strategy9(self.func, self.npar, npop, sfactor, xprob))
xpar = numpy.asarray(xpar)
if step is None:
step = xpar * 1.2 + 1.2
factor = 10
self.polytope = \
SimplexRandom(func, npop, xpar, xmin, xmax, step, seed, factor)
self.local_opt = self.ncores_nm.algo
return
def __call__(self, maxnfev, ftol):
random.seed(self.seed)
mypop = self.polytope
npop_1 = self.npop - 1
while self.nfev[0] < maxnfev:
for pop_index in range(self.npop):
key = self.calc_key([pop_index])
# trial = self.all_strategies(pop_index)
trial = self.all_strategies(key[0])
if trial[-1] < mypop[npop_1][-1]:
mypop[npop_1] = trial[1:]
self.polytope.sort()
if self.check_convergence(mypop, ftol, 0):
break
best_vertex = mypop[0]
best_par = best_vertex[:-1]
best_val = best_vertex[-1]
return self.nfev[0], best_val, best_par
def all_strategies(self, key):
rand, index = self.key2.parse(key)
random.seed(rand)
mypop = self.polytope
best_trial = self.strategies[0](mypop, index)
for ii in range(1, len(self.strategies)):
trial = self.strategies[ii](mypop, index)
if trial[-1] < best_trial[-1]:
best_trial = trial
if best_trial[-1] < mypop[0][-1]:
best_trial = self.apply_local_opt(best_trial, index)
return best_trial
def apply_local_opt(self, arg, index):
local_opt = self.local_opt[index % len(self.local_opt)]
result = local_opt(self.func, arg[1:-1], self.xmin, self.xmax)
tmp = numpy.append(result[0], result[2])
result = numpy.append(tmp, result[1])
return result
def calc_key(self, indices, start=0, end=65536):
result = numpy.empty(len(indices), dtype=numpy.int64)
for ii, index in enumerate(indices):
rand = random.randint(start, end)
result[ii] = self.key2.calc(rand, index)
return result
def check_convergence(self, mypop, ftol, npar):
fval_std = numpy.std([col[-1] for col in mypop])
if fval_std < ftol:
return True
return False