def simplex_init_new_generation(self, gen):
'''It will run the simplex method every simplex_interval
generation with a fracitonal step given by simple_step
on the simplex_n*n_pop best individuals.
'''
print 'Inits new generation'
if gen%self.simplex_interval == 0:
spread = array(self.trial_vec).max(0) - array(self.trial_vec).min(0)
indices = argsort(self.fom_vec)
n_ind = int(self.n_pop*self.simplex_n)
if n_ind == 0:
n_ind = 1
# Apply the simplex to a simplex_n memebers (0-1)
for index1 in indices[:n_ind]:
self.text_output('Starting simplex run for member: %d'%index1)
mem = self.pop_vec[index1].copy()
mem_fom = self.fom_vec[index1]
simp = Simplex(self.calc_fom, mem, spread*self.simplex_step)
new_vec, err, iter = simp.minimize(epsilon = self.best_fom/self.simplex_rel_epsilon, maxiters = self.simplex_max_iter)
if self.use_boundaries:
# Check so that the parameters lie indside the bounds
ok = bitwise_and(self.par_max > new_vec, self.par_min < new_vec)
# If not inside make a random re-initilazation of that parameter
new_vec = where(ok, new_vec, random.rand(self.n_dim)*\
(self.par_max - self.par_min) + self.par_min)
new_fom = self.calc_fom(new_vec)
if new_fom < mem_fom:
self.pop_vec[index1] = new_vec.copy()
self.fom_vec[index1] = new_fom
if new_fom < self.best_fom:
self.best_fom = new_fom
self.best_vec = new_vec.copy()
self.new_best = True
def simplex_old_init_new_generation(self, gen):
'''It will run the simplex method every simplex_interval
generation with a fracitonal step given by simple_step
on the best indivual as well a random fraction of simplex_n individuals.
'''
print 'Inits new generation'
if gen%self.simplex_interval == 0:
spread = array(self.trial_vec).max(0) - array(self.trial_vec).min(0)
simp = Simplex(self.calc_fom, self.best_vec, spread*self.simplex_step)
print 'Starting simplex run for best vec'
new_vec, err, iter = simp.minimize(epsilon = self.best_fom/self.simplex_rel_epsilon, maxiters = self.simplex_max_iter)
print 'FOM improvement: ', self.best_fom - err
if self.use_boundaries:
# Check so that the parameters lie indside the bounds
ok = bitwise_and(self.par_max > new_vec, self.par_min < new_vec)
# If not inside make a random re-initilazation of that parameter
new_vec = where(ok, new_vec, random.rand(self.n_dim)*\
(self.par_max - self.par_min) + self.par_min)
new_fom = self.calc_fom(new_vec)
if new_fom < self.best_fom:
self.best_fom = new_fom
self.best_vec = new_vec
self.pop_vec[0] = new_vec
self.fom_vec[0] = self.best_fom
self.new_best = True
# Apply the simplex to a simplex_n memebers (0-1)
for index1 in random_mod.sample(xrange(len(self.pop_vec)),
int(len(self.pop_vec)*self.simplex_n)):
print 'Starting simplex run for member: ', index1
mem = self.pop_vec[index1]
mem_fom = self.fom_vec[index1]
simp = Simplex(self.calc_fom, mem, spread*self.simplex_step)
new_vec, err, iter = simp.minimize(epsilon = self.best_fom/self.simplex_rel_epsilon, maxiters = self.simplex_max_iter)
if self.use_boundaries:
# Check so that the parameters lie indside the bounds
ok = bitwise_and(self.par_max > new_vec, self.par_min < new_vec)
# If not inside make a random re-initilazation of that parameter
new_vec = where(ok, new_vec, random.rand(self.n_dim)*\
(self.par_max - self.par_min) + self.par_min)
new_fom = self.calc_fom(new_vec)
if new_fom < mem_fom:
self.pop_vec[index1] = new_vec
self.fom_vec[index1] = new_fom
if new_fom < self.best_fom:
self.best_fom = new_fom
self.best_vec = new_vec
self.new_best = True
