def _load_init_population(self, intcodedir, size=None, copy=True):
# make sure we are at the beginning of experiment
assert self._istep == 0, 'initialization only allowed at the beginning'
# make sure size <= population size
if size is not None:
assert size <= self._popsize, f'size {size} too big for population of size {self._popsize}'
else:
size = self._popsize
# load codes
init_population, genealogy = utils.load_codes2(intcodedir, size,
self._random_generator)
init_population = init_population.reshape(
(len(init_population), *self._code_shape))
# fill the rest of population if size==len(codes) < population size
if len(init_population) < self._popsize:
remainder_size = self._popsize - len(init_population)
remainder_pop, remainder_genealogy = \
self._mate(init_population, genealogy, np.ones(len(init_population)), remainder_size)
remainder_pop, remainder_genealogy = self._mutate(
remainder_pop, remainder_genealogy)
init_population = np.concatenate((init_population, remainder_pop))
genealogy = genealogy + remainder_genealogy
# apply
self._init_population = init_population
self._init_population_fns = genealogy # a list of '*.npy' file names
self._curr_samples = self._init_population.copy()
self._genealogy = [f'[init]{g}' for g in genealogy]
# no update for idc, idx, ids because popsize unchanged
if copy:
self._copy_init_population()
if self._generator.loaded:
self._prepare_images()
def _load_init_code(self, initcodedir, copy=True):
# make sure we are at the beginning of experiment
assert self._istep == 0, 'initialization only allowed at the beginning'
init_code, init_codefns = utils.load_codes2(initcodedir, 1,
self._random_generator)
self._curr_center = init_code.reshape(self._code_shape)
self._init_codefn = init_codefns[0]
self._best_code = self._curr_center.copy()
if copy:
self._copy_init_code()
('caffe-net', 'fc8', 1), ]
Optim_arr = ["Genetic", "CholCMA"]
#%% Genetic Algorithm Parameters AND CMA-ES Parameters
population_size = 40
mutation_rate = 0.25
mutation_size = 0.75
kT_multiplier = 2
n_conserve = 10
parental_skew = 0.75
# Genetic(population_size, mutation_rate, mutation_size, kT_multiplier, recorddir,
# parental_skew=0.5, n_conserve=0)
code_length = 4096
init_sigma = 3
Aupdate_freq = 10
# use the mean code of the texture patterns as the initcode
codes, _ = utils.load_codes2(initcodedir, 40)
initcode = np.mean(codes, axis=0, keepdims=True)
# CholeskyCMAES(recorddir=recorddir, space_dimen=code_length, init_sigma=init_sigma,
# Aupdate_freq=Aupdate_freq, init_code=np.zeros([1, code_length]))
#%%
from time import time
import matplotlib.pylab as plt
for unit in unit_arr[:]:
savedir = join(recorddir, "optim_cmp", "%s_%s_%d" % (unit[0], unit[1], unit[2]))
os.makedirs(savedir, exist_ok=True)
for Optim_str in Optim_arr:
for triali in range(10):
t0 = time()
if Optim_str == "Genetic":
optim = Genetic(population_size, mutation_rate, mutation_size, kT_multiplier, recorddir,