def __init__(self,
network,
eps=0.07,
alpha=None,
bounds=None,
loss_fn=None):
super(GradientMethod, self).__init__()
self._network = check_model('network', network, Cell)
self._eps = check_value_positive('eps', eps)
self._dtype = None
if bounds is not None:
self._bounds = check_param_multi_types('bounds', bounds,
[list, tuple])
for b in self._bounds:
_ = check_param_multi_types('bound', b, [int, float])
else:
self._bounds = bounds
if alpha is not None:
self._alpha = check_value_positive('alpha', alpha)
else:
self._alpha = alpha
if loss_fn is None:
loss_fn = SoftmaxCrossEntropyWithLogits(is_grad=False,
sparse=False)
with_loss_cell = WithLossCell(self._network, loss_fn)
self._grad_all = GradWrapWithLoss(with_loss_cell)
self._grad_all.set_train()
def __init__(self,
network,
eps=1e-5,
bounds=(0.0, 1.0),
is_targeted=True,
nb_iter=150,
search_iters=30,
loss_fn=None,
sparse=False):
super(LBFGS, self).__init__()
self._network = check_model('network', network, Cell)
self._eps = check_value_positive('eps', eps)
self._is_targeted = check_param_type('is_targeted', is_targeted, bool)
self._nb_iter = check_int_positive('nb_iter', nb_iter)
self._search_iters = check_int_positive('search_iters', search_iters)
if loss_fn is None:
loss_fn = SoftmaxCrossEntropyWithLogits(is_grad=False,
sparse=False)
with_loss_cell = WithLossCell(self._network, loss_fn)
self._grad_all = GradWrapWithLoss(with_loss_cell)
self._dtype = None
self._bounds = check_param_multi_types('bounds', bounds, [list, tuple])
self._sparse = check_param_type('sparse', sparse, bool)
for b in self._bounds:
_ = check_param_multi_types('bound', b, [int, float])
box_max, box_min = bounds
if box_max < box_min:
self._box_min = box_max
self._box_max = box_min
else:
self._box_min = box_min
self._box_max = box_max
def __init__(self, network, eps=0.3, eps_iter=0.1, bounds=(0.0, 1.0), nb_iter=5,
loss_fn=None):
super(IterativeGradientMethod, self).__init__()
self._network = check_model('network', network, Cell)
self._eps = check_value_positive('eps', eps)
self._eps_iter = check_value_positive('eps_iter', eps_iter)
self._nb_iter = check_int_positive('nb_iter', nb_iter)
self._bounds = None
if bounds is not None:
self._bounds = check_param_multi_types('bounds', bounds, [list, tuple])
for b in self._bounds:
_ = check_param_multi_types('bound', b, [int, float])
if loss_fn is None:
self._loss_grad = network
else:
self._loss_grad = GradWrapWithLoss(WithLossCell(self._network, loss_fn))
self._loss_grad.set_train()