MaxAcquisitionPolicy

class probnum.quad.solvers.policies.MaxAcquisitionPolicy(batch_size, acquisition_func, n_restarts)

Bases: Policy

Policy that maximizes an acquisition function with an optimizer.

Parameters:
  • batch_size (IntLike) – Size of batch of nodes when calling the policy once (must be equal to 1).

  • acquisition_func (AcquisitionFunction) – The acquisition function.

  • n_restarts (IntLike) – The number of times the optimizer is being restarted.

Raises:

ValueError – If batch_size is not 1, or if n_restarts is too small.

Notes

The policy uses SciPy’s ‘Nelder-Mead’ optimizer when gradients are unavailable. This is the current standard setting since probnum does not provide gradients yet.

Attributes Summary

requires_rng

Whether the policy requires a random number generator when called.

Methods Summary

__call__(bq_state, rng)

Find nodes according to the policy.

Attributes Documentation

requires_rng

Methods Documentation

__call__(bq_state, rng)[source]

Find nodes according to the policy.

Parameters:
  • bq_state (BQState) – State of the BQ belief.

  • rng (Generator | None) – A random number generator.

Returns:

shape=(batch_size, input_dim) – Nodes found according to the policy.

Return type:

nodes