CombinationSparseGrid

class pyapprox.surrogates.CombinationSparseGrid(num_vars, basis_type='barycentric')

Bases: SubSpaceRefinementManager

Adaptive sparse grid that uses the combination technique.

Methods Summary

__call__(samples[, return_grad])	config values are ignored.
add_new_subspaces(new_subspace_indices)
evaluate_at_interrogation_samples()	Evaluate the sparse grid at self.canonical_interrogation_samples.
evaluate_using_all_data(samples)	Evaluate sparse grid using all subspace indices including active subspaces.
get_samples()
get_subspace_samples(subspace_index, ...)	Must be implemented by derived class This function should only be called when updating grid not interogating grid
moments()
moments_(smolyak_coefficients)
refine_and_add_new_subspaces(...)
save(filename)
set_interrogation_samples(samples)	Set samples which are used to evaluate a sparse grid repeatedly.
set_univariate_rules(univariate_quad_rule[, ...])
setup(function, config_variables_idx, ...[, ...])

Methods Documentation

__call__(samples, return_grad=False)

config values are ignored. The sparse grid just returns its best approximation of the highest fidelity model. TODO: consider enforcing that samples do not have configure variables

add_new_subspaces(new_subspace_indices)

evaluate_at_interrogation_samples()

Evaluate the sparse grid at self.canonical_interrogation_samples.

Note, this fuction only uses subspaces which are not active

evaluate_using_all_data(samples)

Evaluate sparse grid using all subspace indices including active subspaces. __call__ only uses subspaces which are not active

get_samples()

get_subspace_samples(subspace_index, unique_poly_indices)

Must be implemented by derived class This function should only be called when updating grid not interogating grid

moments()

moments_(smolyak_coefficients)

refine_and_add_new_subspaces(best_active_subspace_index)

save(filename)

set_interrogation_samples(samples)

Set samples which are used to evaluate a sparse grid repeatedly. If provided each time a subspace is added the subspace is evaluated at these points so that when self.evaluate_at_interrogation_samples is called no major computations are required. Note the reduced time complexity requires more storage

Parameters:

samplesnp.ndarray (num_vars) or (num_vars-num_config_vars)

Samples at which to evaluate the sparae grid. If config values are provided they are ignored.

set_univariate_rules(univariate_quad_rule, max_level=2)

setup(function, config_variables_idx, refinement_indicator, admissibility_function, univariate_growth_rule, univariate_quad_rule, variable_transformation=None, config_var_trans=None, cost_function=None, work_qoi_index=None, unique_quadrule_indices=None, verbose=0)