tf.keras.optimizers.Adamax( learning_rate=0.001, beta_1=0.9, beta_2=0.999, epsilon=1e-07, name="Adamax", **kwargs )
Optimizer that implements the Adamax algorithm.
It is a variant of Adam based on the infinity norm. Default parameters follow those provided in the paper. Adamax is sometimes superior to adam, specially in models with embeddings.
m = 0 # Initialize initial 1st moment vector v = 0 # Initialize the exponentially weighted infinity norm t = 0 # Initialize timestep
The update rule for parameter
w with gradient
described at the end of section 7.1 of the paper:
t += 1 m = beta1 * m + (1 - beta) * g v = max(beta2 * v, abs(g)) current_lr = learning_rate / (1 - beta1 ** t) w = w - current_lr * m / (v + epsilon)
Adam, the epsilon is added for numerical stability
(especially to get rid of division by zero when
v_t == 0).
In contrast to
Adam, the sparse implementation of this algorithm
(used when the gradient is an IndexedSlices object, typically because of
tf.gather or an embedding lookup in the forward pass) only updates
variable slices and corresponding
v_t terms when that part of
the variable was used in the forward pass. This means that the sparse
behavior is contrast to the dense behavior (similar to some momentum
implementations which ignore momentum unless a variable slice was actually
Tensor, floating point value, or a schedule that is a
tf.keras.optimizers.schedules.LearningRateSchedule. The learning rate.
"clipnorm"(float) clips gradients by norm;
"clipvalue"(float) clips gradients by value.