AbstractBayesianOED

class pyapprox.expdesign.AbstractBayesianOED(ndesign_candidates, obs_fun, noise_std, prior_variable, out_quad_opts, in_quad_opts, nprocs=1, max_ncollected_obs=2, ndata_per_candidate=1, data_risk_fun=<function oed_data_expectation>)

Bases: ABC

Base Bayesian OED class

Methods Summary

compute_expected_utility(...[, return_all])
compute_utilities(ncandidates, ...)
populate()
select_design(collected_design_indices, ...)	Update an experimental design.
set_collected_design_indices(indices)
update_design([return_all, nnew])

Methods Documentation

abstract compute_expected_utility(collected_design_indices, new_design_indices, return_all=False)

compute_utilities(ncandidates, collected_design_indices, new_indices, return_all)

abstract populate()

select_design(collected_design_indices, nnew, return_all)

Update an experimental design.

Parameters:

collected_design_indicesnp.ndarray (nobs)

The indices into the qoi vector associated with the collected observations

Returns:

utility_valsnp.ndarray (ncandidates)

The utility vals at the candidate design samples. If the candidate sample is already in collected design then the utility value will be set to -np.inf

selected_indexinteger

The index of the best design, i.e. the largest utility

resultsdict

Dictionary of useful data used to compute expected utility At a minimum it has the keys [“utilties”, “evidences”, “weights”]

set_collected_design_indices(indices)

update_design(return_all=False, nnew=1)