A LearningRateSchedule that uses a polynomial decay schedule

  end_learning_rate = 1e-04,
  power = 1,
  cycle = FALSE,
  name = NULL



A scalar float32 or float64 Tensor or an R number. The initial learning rate.


A scalar int32 or int64 Tensor or an R number. Must be positive. See the decay computation above.


A scalar float32 or float64 Tensor or an R number. The minimal end learning rate.


A scalar float32 or float64 Tensor or an R number. The power of the polynomial. Defaults to linear, 1.0.


A boolean, whether or not it should cycle beyond decay_steps.


For backwards and forwards compatibility


String. Optional name of the operation. Defaults to 'PolynomialDecay'.


It is commonly observed that a monotonically decreasing learning rate, whose degree of change is carefully chosen, results in a better performing model. This schedule applies a polynomial decay function to an optimizer step, given a provided initial_learning_rate, to reach an end_learning_rate in the given decay_steps.

It requires a step value to compute the decayed learning rate. You can just pass a TensorFlow variable that you increment at each training step.

The schedule is a 1-arg callable that produces a decayed learning rate when passed the current optimizer step. This can be useful for changing the learning rate value across different invocations of optimizer functions. It is computed as:

decayed_learning_rate <- function(step) {
  step <- min(step, decay_steps)
  ((initial_learning_rate - end_learning_rate) *
      (1 - step / decay_steps) ^ (power)
    ) + end_learning_rate

If cycle is TRUE then a multiple of decay_steps is used, the first one that is bigger than step.

decayed_learning_rate <- function(step) {
  decay_steps <- decay_steps * ceiling(step / decay_steps)
  ((initial_learning_rate - end_learning_rate) *
      (1 - step / decay_steps) ^ (power)
    ) + end_learning_rate

You can pass this schedule directly into a keras Optimizer as the learning_rate.

Example: Fit a model while decaying from 0.1 to 0.01 in 10000 steps using sqrt (i.e. power=0.5):

starter_learning_rate <- 0.1
end_learning_rate <- 0.01
decay_steps <- 10000
learning_rate_fn <- learning_rate_schedule_polynomial_decay(
  starter_learning_rate, decay_steps, end_learning_rate, power = 0.5)

model %>%
  compile(optimizer = optimizer_sgd(learning_rate = learning_rate_fn),
          loss = 'sparse_categorical_crossentropy',
          metrics = 'accuracy')

model %>% fit(data, labels, epochs = 5)