tfp.experimental.sequential.geometric_cooling_schedule

View source on GitHub

Defines a cooling schedule following a geometric sequence.

This returns a function f such that

f(iteration) = cooling_fraction_per_k_iterations**(iteration / k)

cooling_fraction_per_k_iterations float Tensor ratio by which the original value should be scaled once k iterations have been completed.
k int Tensor number of iterations used to define the schedule.

f Python callable representing the cooling schedule.