def __call__(self, x):
"""This is the objective function to be minimized with Bayesian optimization.
This function changes the training hyperparameters as provided by x and trains a model. This model
is then evaluated. The selected optimization criteria is returned by the evaluation function, and will be
provided as scalar output.
Args:
x(np.array): 2-D array with a where every row is a list of options.
Returns:
np.array: 2-D array with the cost of the objective per row.
"""
self.opt.script = "generative"
y = np.zeros((x.shape[0], 1))
for i in range(x.shape[0]):
# Overwrite default options with options provided by the bayesian optimization
self.opt = self._set_options(self.opt, x, i)
self.opt.mode = 'train'
print_flags(self.opt)
train(self.opt)
self.opt.mode = 'test'
y[i], stats = test(self.opt)
self._store_stats(stats, vars(self.opt), y[i])
if self.opt.two_test:
self.opt.use_test_set = 1
res, stats = test(self.opt)
self._store_stats(stats, vars(self.opt), res)
self.opt.use_test_set = 0
self.iter += 1
return y
def main():
opt, parser = parse_arguments()
opt = predefined(opt)
print_flags(opt)
if not osp.isdir(opt.out_folder):
os.makedirs(opt.out_folder)
if opt.script == 'generative':
if opt.mode == 'train':
train(opt)
elif opt.mode == 'test':
test(opt)
elif opt.mode == 'generate':
generate_data(opt)
elif opt.mode == 'novelty':
novelty(opt)
elif opt.mode == 'qualitative':
qualitative(opt)
else:
raise UnknownArgumentError(
"--mode not recognized, please choose: [train, test, generate, qualitative, novelty]."
)
elif opt.script == 'bayesopt':
optimize_bayesian(opt, parser)
elif opt.script == 'grid':
run_grid(opt, parser)
else:
raise UnknownArgumentError(
"--script not recognized, please choose: [generative, bayesopt, grid]."
)
def re_test(self, run_ids):
self.opt.script = "generative"
self.opt.mode = 'test'
for run in run_ids:
self.iter = run
self.opt.save_suffix = "{}_{}".format(self.opt.bayes_mode,
self.iter)
y, stats = test(self.opt)
self._store_stats(stats, vars(self.opt), y)
if self.opt.two_test:
self.opt.use_test_set = 1
res, stats = test(self.opt)
self._store_stats(stats, vars(self.opt), res)
self.opt.use_test_set = 0