UBay constraints

UBayFS

class UBayconstraint.UBayconstraint(rho, A=None, b=None, block_matrix=None, block_list=None, constraint_types=None, constraint_vars=None, num_elements=None)

This class initializes user-defined constraints.

Parameters:

  • rho (<numpy array>) – Vector of regularization strenghts for the defined constraints.

  • A (<numpy array>) – Matrix describing constraints. Left side of the equation system. Default: A=None

  • b (<numpy array>) – 1-d dimensional array defining the right sight of the equation system. Default: b=None

  • block_matrix (<numpy array>) – Matrix describing the block assignment for each feature. If no block-structure given, block_matrix is a diagonal-unity matrix. Default : block_matrix=None.

  • block_list (<list>) – List describing the block assignment for each feature, if block structure is present. Default : block_list=None.

  • constraint_types (<list> of <strings>) –

    Constraint types. Possible options are:

    • ”max_size” : maximal number of features that shall be selected

    • ”must_link” : the defined feature set must be selected together

    • ”cannot_link” : the defined feature set must not be selected together

  • constraint_vars (<list> of <int> or <lists>) – For each constraint_type, define features sets that are concerned. Default : `constraint_vars=None.

  • num_elements (<int>) – Total number of features. Default :num_elements=None.

get_constraints()

Get the constraints.

Return type:

A dictionary including A, b, and block_matrix.

get_dimensions()

Get the dimensions of the constraint matrix A and the block matrix.

Return type:

…

get_maxsize()

Get the right side (b) of the max size constraint.

Return type:

An integer.

group_admissibility(state, log=True)

Evaluate the value of the admissibility function ‘kappa’ for a group of constraints (with a common block)-

Parameters:

  • state (<np.array>) – 1-dimensional binary array describing a feature set. 1: feature selected, 0: feature not selected.

  • log (<boolean>) – Indicates whether the admissibility should be returned on log scale.

Return type:

An admissibility value <float>.