BayesianSequentialOED

class pyapprox.expdesign.BayesianSequentialOED(obs_process)

Bases: AbstractBayesianOED

Compute sequential optimal experimental designs that collect data and use this to inform the choice of subsequent design locations.

Methods Summary

compute_importance_weights()	Compute the importance weights used in the computation of the expected utility that acccount for the fact we want to use the current posterior as the prior in the utility formula.
set_collected_design_indices(indices)	Set the initial design indices and collect data at the corresponding design points.
update_observations(new_obs)	Store the newly collected obsevations which will dictate the next design point.

Methods Documentation

compute_importance_weights()

Compute the importance weights used in the computation of the expected utility that acccount for the fact we want to use the current posterior as the prior in the utility formula.

set_collected_design_indices(indices)

Set the initial design indices and collect data at the corresponding design points.

Parameters:

indicesnp.ndarray (nindices, 1)

The indices corresponding to an initial design

update_observations(new_obs)

Store the newly collected obsevations which will dictate the next design point.

Parameters:

new_obsnp.ndarray (1, nnew_obs)

The new observations

Notes

self.inner_importance_weights contains likelihood vals/evidence at in_samples self.in_weights is the prior quadrature weights which for random samples drawn from prior is just 1/N and for Gauss Quadrature is the quadrature rule weights.

Similarly for self.outer_importance_weights