Mava icon indicating copy to clipboard operation
Mava copied to clipboard

[FEATURE] piecewise linear epsilon for exploration

Open nima-siboni opened this issue 2 years ago • 0 comments

Feature

A piecewise linear scheduler for epsilon. With piecewise linear scheduler the user can increase and decrease the epsilon over the desired time intervals.

Proposal

Creating a new class for similar to LinearTimestepScheduler

Testing

The returned value of the epsilon should be tested:

  • The returned value right after creation of an instance
  • The returned value after updating to a timestep before the smallest timestep in the given list, i.e. the initial fallback value
  • The returned value after updating to a timestep beyond the largest timestep in the given list, i.e. the final fallback value
  • The returned value interpolating between the given timesteps and epsilons

Definition of done

A class is implemented and all the tests are passed.

nima-siboni avatar Apr 11 '22 08:04 nima-siboni