hex.genmodel.algos.gam

Class NBSplinesTypeI

java.lang.Object

hex.genmodel.algos.gam.NBSplinesTypeI

All Implemented Interfaces:

java.io.Serializable

public class NBSplinesTypeI
extends java.lang.Object
implements java.io.Serializable

I implemented the spline described in Section III of doc in the GitHub issue https://github.com/h2oai/h2o-3/issues/7261. Any reference to doc I in the code refer to the one here with the http link. The recursive formula in equation 5 is used. It is implemented as a binary tree with current node with order m and two child nodes with order m-1.

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	NBSplinesTypeI.MSplineBasis This class describes a M spline using the recursive formula in equation 5 of the doc.

Field Summary

Fields

Modifier and Type	Field and Description
double	_commonConst
int	_nKnots
double[][]	_nodeCoeffs
int	_order
int	_totBasisFuncs

Constructor Summary

Constructors

Constructor and Description
NBSplinesTypeI(double[] knots, int order, int basisIndex, int numKnotInt) Generate SBSpline Type I in Section III of doc I

Method Summary

Modifier and Type	Method and Description
static double[][]	extractCoeffs(NBSplinesTypeI root, int basisIndex, double parentConst) extract coefficients of NBSplineType I in the process of generating penalty matrix in Section VI of doc.
static void	extractNBSplineCoeffs(NBSplinesTypeI root, int order, double[] coeffParent, double constParent, int basisIndex) Given root spline, this method will extract the coefficients of spline as described in Section VI in order to generate the penalty matrix using recursion.
static NBSplinesTypeI	formBasisDeriv(double[] knots, int order, int basisIndex, int numKnotInt) Given an basis function index, order and knot sequence with duplication over the input range of interest, we build the whole binary tree for NBSplineType I for Mi,k(t) down to the lowest level with splines of order 1 as in Mi,1(t).
protected static NBSplinesTypeI.MSplineBasis[]	genBasisFunctions(int totBasisFuncs, int order, double[] knots)
void	setConstNChildCoeffs(double[] knots, int basisIndex)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

_order

public final int _order

_nodeCoeffs

public double[][] _nodeCoeffs

_commonConst

public double _commonConst

_totBasisFuncs

public final int _totBasisFuncs

_nKnots

public final int _nKnots

Constructor Detail

NBSplinesTypeI

public NBSplinesTypeI(double[] knots,
                      int order,
                      int basisIndex,
                      int numKnotInt)

Generate SBSpline Type I in Section III of doc I

Parameters:

knots - : knots that span the whole input range of interest with no duplication

order - : order of spline

basisIndex - : offset added to basis function index, recall lower order splines have index i and i+1 equation 5 of doc

numKnotInt - : integer denoting knot intervals over which polynomial coefficients are defined

Method Detail

genBasisFunctions

protected static NBSplinesTypeI.MSplineBasis[] genBasisFunctions(int totBasisFuncs,
                                                                 int order,
                                                                 double[] knots)

setConstNChildCoeffs

public void setConstNChildCoeffs(double[] knots,
                                 int basisIndex)

extractNBSplineCoeffs

public static void extractNBSplineCoeffs(NBSplinesTypeI root,
                                         int order,
                                         double[] coeffParent,
                                         double constParent,
                                         int basisIndex)

Given root spline, this method will extract the coefficients of spline as described in Section VI in order to generate the penalty matrix using recursion. This is actually follows recursion of TypeII

extractCoeffs

public static double[][] extractCoeffs(NBSplinesTypeI root,
                                       int basisIndex,
                                       double parentConst)

extract coefficients of NBSplineType I in the process of generating penalty matrix in Section VI of doc.

formBasisDeriv

public static NBSplinesTypeI formBasisDeriv(double[] knots,
                                            int order,
                                            int basisIndex,
                                            int numKnotInt)

Given an basis function index, order and knot sequence with duplication over the input range of interest, we build the whole binary tree for NBSplineType I for Mi,k(t) down to the lowest level with splines of order 1 as in Mi,1(t). This is done using recursion following equation 5.

Parameters:

knots - : knot sequence with duplication

order - : order (k) of spline to build

basisIndex - : offset to added in order to address spline of lower order

numKnotInt - : length of knots with duplication over the whole range of input of interest.

Returns:

NBSplinesTypeI for spline Mi,k(t)