BayesianSequentialKLOED
- class pyapprox.expdesign.BayesianSequentialKLOED(ndesign_candidates, obs_fun, noise_std, prior_variable, out_quad_opts, in_quad_opts, obs_process=None, nprocs=1, max_ncollected_obs=2)[source]
Bases:
BayesianSequentialOED,BayesianBatchKLOEDCompute closed-loop OED my maximizing KL divergence between the prior and posterior.
Methods Summary
compute_expected_utility(...[, return_all])Compute the expected utility.
Methods Documentation
- compute_expected_utility(collected_design_indices, new_design_indices, return_all=False)[source]
Compute the expected utility. Using the current posterior as the new prior.
- Parameters:
- collected_design_indicesnp.ndarray (nobs)
The indices into the qoi vector associated with the collected observations
- new_design_indicesnp.ndarray (nnew_obs)
The indices into the qoi vector associated with new design locations under consideration
Notes
Passing None for collected_design_indices will ensure only obs at new_design indices is used to evaluate likelihood the data at collected indices is incoroporated into the inner and outer loop weights