def __call__(self, w):
maximum_weight = K.max(K.abs(w))
w = w / (K.epsilon() + maximum_weight) # On [-1,1].
signs = K.sign(w)
unsigned_w = K.abs(w) # On [0,1].
edge_scaled_w_unsigned = unsigned_w * (
1. - self.min_value) + self.min_value # On [min_value,1].
edge_scaled_w = signs * edge_scaled_w_unsigned # On [-1,-min_value] U [min_value,1].
return edge_scaled_w
def __call__(self, x):
regularization = 0.
if self.l1:
regularization += K.sum(self.l1 * K.abs(x))
if self.l2:
regularization += K.sum(self.l2 * K.square(x))
return regularization
def __call__(self, x):
regularization = 0.
if self.l1:
regularization += K.sum(self.l1 * K.abs(x))
if self.l2:
regularization += K.sum(self.l2 * K.square(x))
return regularization
def get_updates(self, loss, params):
grads = self.get_gradients(loss, params)
self.updates = [K.update_add(self.iterations, 1)]
lr = self.lr
if self.initial_decay > 0:
lr *= (1. /
(1. + self.decay * K.cast(self.iterations, K.dtype(self.decay))))
t = K.cast(self.iterations, K.floatx()) + 1
lr_t = lr / (1. - K.pow(self.beta_1, t))
shapes = [K.int_shape(p) for p in params]
# zero init of 1st moment
ms = [K.zeros(shape) for shape in shapes]
# zero init of exponentially weighted infinity norm
us = [K.zeros(shape) for shape in shapes]
self.weights = [self.iterations] + ms + us
for p, g, m, u in zip(params, grads, ms, us):
m_t = (self.beta_1 * m) + (1. - self.beta_1) * g
u_t = K.maximum(self.beta_2 * u, K.abs(g))
p_t = p - lr_t * m_t / (u_t + self.epsilon)
self.updates.append(K.update(m, m_t))
self.updates.append(K.update(u, u_t))
new_p = p_t
# Apply constraints.
if getattr(p, 'constraint', None) is not None:
new_p = p.constraint(new_p)
self.updates.append(K.update(p, new_p))
return self.updates
def call(self, inputs, mask=None):
pos = K.relu(inputs)
if K.backend() == 'theano':
neg = (K.pattern_broadcast(self.alpha, self.param_broadcast) *
(inputs - K.abs(inputs)) * 0.5)
else:
neg = -self.alpha * K.relu(-inputs)
return pos + neg
def call(self, inputs, mask=None):
pos = K.relu(inputs)
if K.backend() == 'theano':
neg = (K.pattern_broadcast(self.alpha, self.param_broadcast) *
(inputs - K.abs(inputs)) * 0.5)
else:
neg = -self.alpha * K.relu(-inputs)
return pos + neg
def dice_coef(y_true, y_pred, smooth=1e-5):
"""
Dice = (2*|X & Y|)/ (|X|+ |Y|)
= 2*sum(|A*B|)/(sum(A^2)+sum(B^2))
ref: https://arxiv.org/pdf/1606.04797v1.pdf
THIS IS NOT DICE BUT ...
"""
intersection = K.sum(K.abs(y_true * y_pred), axis=-1)
return (2. * intersection + smooth) / (
K.sum(K.square(y_true), -1) + K.sum(K.square(y_pred), -1) + smooth)
def __call__(self, w):
# First apply DivideByMax.
maximum_weight = K.max(K.abs(w))
w = w / (K.epsilon() + maximum_weight) # On [-1, 1].
# Then apply MinMaxNorm.
norms = K.sqrt(
math_ops.reduce_sum(math_ops.square(w),
axis=self.axis,
keepdims=True))
desired = (self.rate * K.clip(norms, self.min_value, self.max_value) +
(1 - self.rate) * norms)
return w * (desired / (K.epsilon() + norms))
def get_updates(self, loss, params):
grads = self.get_gradients(loss, params)
self.updates = [K.update_add(self.iterations, 1)]
lr = self.lr
if self.initial_decay > 0:
lr = lr * (
1. / # pylint: disable=g-no-augmented-assignment
(1. +
self.decay * K.cast(self.iterations, K.dtype(self.decay))))
t = K.cast(self.iterations, K.floatx()) + 1
lr_t = lr / (1. - K.pow(self.beta_1, t))
shapes = [K.int_shape(p) for p in params]
# zero init of 1st moment
ms = [K.zeros(shape) for shape in shapes]
# zero init of exponentially weighted infinity norm
us = [K.zeros(shape) for shape in shapes]
self.weights = [self.iterations] + ms + us
for p, g, m, u in zip(params, grads, ms, us):
m_t = (self.beta_1 * m) + (1. - self.beta_1) * g
u_t = K.maximum(self.beta_2 * u, K.abs(g))
p_t = p - lr_t * m_t / (u_t + self.epsilon)
self.updates.append(K.update(m, m_t))
self.updates.append(K.update(u, u_t))
new_p = p_t
# Apply constraints.
if getattr(p, 'constraint', None) is not None:
new_p = p.constraint(new_p)
self.updates.append(K.update(p, new_p))
return self.updates
def mean_absolute_percentage_error(y_true, y_pred): # Equivalent to MAE, but sometimes easier to interpret. diff = K.abs((y_true - y_pred) / K.clip(K.abs(y_true), K.epsilon(), None)) return 100. * K.mean(diff, axis=-1)
def mean_absolute_error(y_true, y_pred): return K.mean(K.abs(y_pred - y_true), axis=-1)
def diff_loss(y_true, y_pred):
y_true = K.flatten(y_true)
y_pred = K.flatten(y_pred)
diff = K.sum(K.abs(y_true - y_pred))
return diff
def mean_absolute_percentage_error(y_true, y_pred): diff = K.abs((y_true - y_pred) / K.clip(K.abs(y_true), K.epsilon(), None)) return 100. * K.mean(diff, axis=-1)
def mean_absolute_percentage_error(y_true, y_pred): diff = K.abs((y_true - y_pred) / K.clip(K.abs(y_true), K.epsilon(), None)) return 100. * K.mean(diff, axis=-1)
def __call__(self, w):
maximum_weight = K.max(K.abs(w))
return w / (K.epsilon() + maximum_weight)
