Coordinate-based Dimension Reduction¶
The goal of a coordinate based dimension reduction is to identify subsets of variables that are sufficient to explain the behavor of the function; this process is sometimes known as variable screening.
Although in many ways this is a simpler process than subspace based dimension reduction, from the point of view of the code, a coordinate-based dimension reduction is a special case of a subspace-based dimension reduction where subspaces are built from columns of the identity matrix.
-
class
psdr.
CoordinateBasedDimensionReduction
[source]¶ Abstract base class for dimension reduction strategies that select variables
-
U
¶ A matrix defining the ‘important’ directions
Returns: Matrix with orthonormal columns defining the important directions in decreasing order of precidence. Return type: np.ndarray (m, n)
-
score
¶ The score associated with each parameter
-
Diagonal Lipschitz Matrix¶
-
class
psdr.
DiagonalLipschitzMatrix
(L=None, **kwargs)[source]¶ Constructs a diagonal Lipschitz matrix
Much like the standard Lipschitz matrix class
psdr.LipschitzMatrix()
, this class computes a Lipschitz matrix except with the constraint the matrix is diagonal.