def nll(y_true, y_pred):
#if variance_logits:
# variance_tensor = K.exp(variance_tensor)
return 0.5 * K.mean(
K.log(variance_tensor + epsilon) + K.square(y_true - y_pred) /
(variance_tensor + epsilon))
def beta_nll(y_true, y_pred):
#if variance_logits:
# variance_tensor = K.exp(variance_tensor)
beta_sigma_sq = K.stop_gradient(K.pow(variance_tensor, beta))
return 0.5 * K.mean(
beta_sigma_sq *
(K.log(variance_tensor + epsilon) + K.square(y_true - y_pred) /
(variance_tensor + epsilon)))
def add_gradient_penalty(model, lambda_coeff=0.5, penalty_type="two-sided"):
term = K.gradients(K.sum(model.output, axis=1), model.input)
term = K.square(term)
if penalty_type == "two-sided":
penalty = (term - 1)**2
elif penalty_type == "one-sided":
penalty = K.max(0, term - 1)
else:
raise ValueError(
"Invalid penalty type {}, valid values are [one-sided, two-sided]".
format(penalty_type))
penalty = lambda_coeff * penalty
penalty = K.in_train_phase(penalty, K.zeros(shape=(1, )))
model.add_loss(penalty)
def brier_score(y_true, y_pred):
"""
Mean squared error on the probabilities.
"""
return K.mean(K.square(y_true - y_pred))
def rbf(self, z):
z = z - self.centroids
z = K.mean(K.square(z), axis=1) / (2.0 * self.length_scale**2)
z = K.exp(-z)
return z
def log_probability(x, mu, sigma):
return NegHalfLog2PI - K.log(sigma) - 0.5 * K.square((x - mu) / sigma)
def log_probability(self, x):
return NegHalfLog2PI - K.log(self.std) - 0.5 * K.square(
(x - self.mean) / self.std)
def probability(x, mu, sigma):
x = K.square((x - mu) / sigma)
return InvSqrt2PI * (1.0 / sigma) * K.exp(-0.5 * x)