da_step()

Contents

da_step()#

liesel.goose.da.da_step(kernel_state, acceptance_prob, time_in_epoch, target_accept=0.8, gamma=0.05, kappa=0.75, t0=10)[source]#

Performs an dual averaging update on a DAKernelState. Returns None and should be called for the side effect on the kernel_state argument.

Parameters:

  • kernel_state (DAKernelState) – A kernel state implementing the DAKernelState protocol.

  • acceptance_prob (float) – The acceptance probability of this MCMC iteration.

  • time_in_epoch (int) – The number of completed MCMC iterations in this epoch.

  • target_accept (float, default: 0.8) – The target acceptance probability.

  • gamma (float, default: 0.05) – The adaptation regularization scale.

  • kappa (float, default: 0.75) – The adaptation relaxation exponent.

  • t0 (int, default: 10) – The adaptation iteration offset.

Return type:

None

Notes

For an introduction to dual averaging, see the blog post by Colin Carroll [1] and the Stan Reference Manual [2].

Contents