SELECTION_LP_THRESHOLD
- selects from all choices of lambda_p
a solution vector with both low cost and low regularization term
Comments
Given multiple solutions of material parameters and their corresponding
costs and regularization terms, this function selects the solution which
has both a low cost and a low regularization term. The selection
procedure depends on a hyperparameter (threshold_factor_cost
), which
influences how sparse the solution should be, and a hyperparameter
(threshold_factor_parameters
), determining below which value parameters
should be discarded.
Input Arguments
lambda_p
(double) - regularization factors
parameters
(double) - multiple solutions of material parameters (every row
corresponds to one solution)
regularizer
(double) - regularization terms of the given solutions
(without considering weighting factor lambda_p)
cost_without_regularizer
(double) - cost function values of the given
solutions minus the regularization terms (considering weighting factor
lambda_p)
threshold_factor_cost
(double) - factor for computing the cost
threshold C^{th}
threshold_factor_parameters
(double) - factor for computing the
parameter threshold theta^{th}
Output Arguments
parameters_select
(double) - selected solution of the material parameters