def __call__(self, x):
regularization = 0
if self.l1:
regularization += model_ops.sum(self.l1 * tf.abs(x))
if self.l2:
regularization += model_ops.sum(self.l2 * tf.square(x))
return regularization
def __call__(self, x):
regularization = 0
if self.l1:
regularization += model_ops.sum(self.l1 * tf.abs(x))
if self.l2:
regularization += model_ops.sum(self.l2 * tf.square(x))
return regularization
def softmax(x):
ndim = get_ndim(x)
if ndim == 2:
return tf.nn.softmax(x)
elif ndim == 3:
e = tf.exp(x - model_ops.max(x, axis=-1, keepdims=True))
s = model_ops.sum(e, axis=-1, keepdims=True)
return e / s
else:
raise ValueError('Cannot apply softmax to a tensor '
'that is not 2D or 3D. '
'Here, ndim=' + str(ndim))
def softmax(x):
ndim = get_ndim(x)
if ndim == 2:
return tf.nn.softmax(x)
elif ndim == 3:
e = tf.exp(x - model_ops.max(x, axis=-1, keepdims=True))
s = model_ops.sum(e, axis=-1, keepdims=True)
return e / s
else:
raise ValueError('Cannot apply softmax to a tensor '
'that is not 2D or 3D. '
'Here, ndim=' + str(ndim))
def cos(x, y):
denom = (
model_ops.sqrt(model_ops.sum(tf.square(x)) *
model_ops.sum(tf.square(y))) + model_ops.epsilon())
return model_ops.dot(x, tf.transpose(y)) / denom
def cos(x, y):
denom = (model_ops.sqrt(
model_ops.sum(tf.square(x)) * model_ops.sum(tf.square(y))) +
model_ops.epsilon())
return model_ops.dot(x, tf.transpose(y)) / denom
def kullback_leibler_divergence(y_true, y_pred):
y_true = model_ops.clip(y_true, model_ops.epsilon(), 1)
y_pred = model_ops.clip(y_pred, model_ops.epsilon(), 1)
return model_ops.sum(y_true * tf.log(y_true / y_pred), axis=-1)
def __call__(self, p):
return p / (model_ops.epsilon() + model_ops.sqrt(
model_ops.sum(tf.square(p), axis=self.axis, keepdims=True)))
def __call__(self, p):
norms = model_ops.sqrt(model_ops.sum(
tf.square(p), axis=self.axis, keepdims=True))
desired = model_ops.clip(norms, 0, self.m)
p *= (desired / (model_ops.epsilon() + norms))
return p
def kullback_leibler_divergence(y_true, y_pred): y_true = model_ops.clip(y_true, model_ops.epsilon(), 1) y_pred = model_ops.clip(y_pred, model_ops.epsilon(), 1) return model_ops.sum(y_true * tf.log(y_true / y_pred), axis=-1)
def __call__(self, p):
return p / (model_ops.epsilon() + model_ops.sqrt(
model_ops.sum(tf.square(p), axis=self.axis, keepdims=True)))
def __call__(self, p):
norms = model_ops.sqrt(
model_ops.sum(tf.square(p), axis=self.axis, keepdims=True))
desired = model_ops.clip(norms, 0, self.m)
p *= (desired / (model_ops.epsilon() + norms))
return p
