hex.gam.GamSplines

Class ThinPlateRegressionUtils

java.lang.Object

hex.gam.GamSplines.ThinPlateRegressionUtils

public class ThinPlateRegressionUtils
extends java.lang.Object

This class contains functions that perform different functions for generating the thin plate regression splines and the polynomial basis functions

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	ThinPlateRegressionUtils.ScaleTPPenalty this class performs scaling on TP penalty matrices that is done in R.

Constructor Summary

Constructors

Constructor and Description
ThinPlateRegressionUtils()

Method Summary

Modifier and Type	Method and Description
static void	addPermutationList(java.util.List<java.lang.Integer[]> onePolyBasis, java.util.List<java.util.List<java.lang.Integer>> permute1Basis)
static int	calculatem(int d) For thin plate regression, given d (number of predictors for a smooth), it will return m where (m-1) is the maximum polynomial degree in the polynomial basis functions.
static int	calculateM(int d, int m)
static boolean	checkFrameRowNA(water.fvec.Frame chk, long rowIndex)
static boolean	checkRowNA(water.fvec.Chunk[] chk, int rowIndex)
static int[][]	convertList2Array(java.util.List<java.lang.Integer[]> list2Convert, int M, int d)
static void	expandCombo(int[] oneList, int[] polyBasis, java.lang.Integer[] basisOrder) Given a combo found by findOnePerm say for d = 5, m = 4, for degree = 1 to m-1 (3 in this case).
static java.lang.String[]	extractColNames(java.lang.String[] src, int srcStart, int destStart, int length)
static void	fillRowArray(water.fvec.NewChunk[] newChk, int colWidth, double[] fillValue)
static void	fillRowOneValue(water.fvec.NewChunk[] newChk, int colWidth, double fillValue)
static java.util.List<java.lang.Integer[]>	findAllPolybasis(java.util.List<java.lang.Integer[]> onePolyBasis) For each list in onePolyBasis, we still need to find all the permutations for that list.
static void	findOnePerm(int totDegree, int[] degreeCombo, int index, java.util.ArrayList<int[]> allCombos, int[] currCombo) For a fixed degree specified as totDegree, specified a set of combination of polynomials to achieve the totDegree.
static void	findPermute(int[] freqMap, java.util.List<java.lang.Integer> prefix, int remaining, java.util.List<java.util.List<java.lang.Integer>> basisPerm)
static java.util.List<java.lang.Integer[]>	findPolyBasis(int d, int m) This method, given number of predictors in the smooth d, number of polynomials in the polynomial basis m, will generate a list of integer array specifying for each predictors the degree that predictor will have.
static int[]	generateOrderFreq(java.lang.Integer[] oneBasis)
static double[][]	generateStarT(double[][] knots, java.util.List<java.lang.Integer[]> polyBasisDegree, double[] gamColMeanRaw, double[] oneOColStd, boolean standardizeTPSmoothers)
static double[][]	genKnotsMultiplePreds(water.fvec.Frame predictVec, GAMModel.GAMParameters parms, int predIndex) Generate knots for thin plate (TP) smoothers.
static java.lang.String	genThinPlateNameStart(GAMModel.GAMParameters parms, int gamColIndex)
static void	mergeCombos(java.util.ArrayList<int[]> oneCombo, java.lang.Integer[] basisOrder, int[] polyBasis, java.util.List<java.lang.Integer[]> polyBasisSet)
static void	setCombo(int[] currCombo, int index, int degreeCount)

Methods inherited from class java.lang.Object

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

Constructor Detail

ThinPlateRegressionUtils

public ThinPlateRegressionUtils()

Method Detail

calculatem

public static int calculatem(int d)

For thin plate regression, given d (number of predictors for a smooth), it will return m where (m-1) is the maximum polynomial degree in the polynomial basis functions. The formula used is m = ceiling of (d+1)/2+1

Parameters:

d - : integer denoting number of predictors for thin plate regression smooth.

Returns:

m : integer denoting the maximum polynomial degree + 1 for polynomial basis function.

calculateM

public static int calculateM(int d,
                             int m)

findPolyBasis

public static java.util.List<java.lang.Integer[]> findPolyBasis(int d,
                                                                int m)

This method, given number of predictors in the smooth d, number of polynomials in the polynomial basis m, will generate a list of integer array specifying for each predictors the degree that predictor will have. For instance, if for a predictor, the degree is 0, a constant 1 is used. If for a particular predictor, the degree is 2, predictor*predictor is used.

Parameters:

d -

m -

Returns:

findAllPolybasis

public static java.util.List<java.lang.Integer[]> findAllPolybasis(java.util.List<java.lang.Integer[]> onePolyBasis)

For each list in onePolyBasis, we still need to find all the permutations for that list. In addition, we need to add the combination for the 0th order as well. For instance, if the list contains {0,0,1}, we need to add to that the lists {0,1,0} and {1,0,0} as well.

Parameters:

onePolyBasis -

Returns:

addPermutationList

public static void addPermutationList(java.util.List<java.lang.Integer[]> onePolyBasis,
                                      java.util.List<java.util.List<java.lang.Integer>> permute1Basis)

findPermute

public static void findPermute(int[] freqMap,
                               java.util.List<java.lang.Integer> prefix,
                               int remaining,
                               java.util.List<java.util.List<java.lang.Integer>> basisPerm)

generateOrderFreq

public static int[] generateOrderFreq(java.lang.Integer[] oneBasis)

mergeCombos

public static void mergeCombos(java.util.ArrayList<int[]> oneCombo,
                               java.lang.Integer[] basisOrder,
                               int[] polyBasis,
                               java.util.List<java.lang.Integer[]> polyBasisSet)

expandCombo

public static void expandCombo(int[] oneList,
                               int[] polyBasis,
                               java.lang.Integer[] basisOrder)

Given a combo found by findOnePerm say for d = 5, m = 4, for degree = 1 to m-1 (3 in this case). The basis poly is {3,2,1}. The returned list of combo are: {{0, 0, 1}, {0, 1, 0}, {0, 0, 2}, {1,0,0}, {0,1,1}, {0,0,3}}. However, we need to convert this list back to the perspective of the predictors. In this case, we have 5 predictors. This function will translate the list from findOnePerm to the perspective of the predictors. For instance, for degree = 1, we need to have one predictive to have degree of 1 and hence the list should be {1, 0, 0, 0, 0}. For degree = 2, we can have one predictor taking degree of 2 or two predictors each taking degree 1. The same applies to the other degrees. Hence, this function will return the following list: {{1, 0, 0, 0, 0}, {0, 2, 0, 0, 0}, {1, 1, 0, 0, 0}, {0, 0, 3, 0, 0}, {1, 2, 0, 0, 0} and {1, 1, 1, 0, 0}}.

Parameters:

oneList -

polyBasis -

basisOrder -

findOnePerm

public static void findOnePerm(int totDegree,
                               int[] degreeCombo,
                               int index,
                               java.util.ArrayList<int[]> allCombos,
                               int[] currCombo)

For a fixed degree specified as totDegree, specified a set of combination of polynomials to achieve the totDegree. For instance, if degreeCombo = {1,2,3} and the totDegree is 1. There is only one way to achieve it by the array {1,0,0}. If totDegree = 2, there are two arrays that will work: {0,1,0} or {2,0,0}. If totDegree = 3, there will be 3 arrays that will work {3,0,0}, {1,1,0}, {0,0,1}.

Parameters:

totDegree - : integer representing degree of polynomial basis

degreeCombo - : degrees allowed for polynomial basis

index -

allCombos -

currCombo -

setCombo

public static void setCombo(int[] currCombo,
                            int index,
                            int degreeCount)

generateStarT

public static double[][] generateStarT(double[][] knots,
                                       java.util.List<java.lang.Integer[]> polyBasisDegree,
                                       double[] gamColMeanRaw,
                                       double[] oneOColStd,
                                       boolean standardizeTPSmoothers)

fillRowOneValue

public static void fillRowOneValue(water.fvec.NewChunk[] newChk,
                                   int colWidth,
                                   double fillValue)

fillRowArray

public static void fillRowArray(water.fvec.NewChunk[] newChk,
                                int colWidth,
                                double[] fillValue)

checkRowNA

public static boolean checkRowNA(water.fvec.Chunk[] chk,
                                 int rowIndex)

checkFrameRowNA

public static boolean checkFrameRowNA(water.fvec.Frame chk,
                                      long rowIndex)

genThinPlateNameStart

public static java.lang.String genThinPlateNameStart(GAMModel.GAMParameters parms,
                                                     int gamColIndex)

extractColNames

public static java.lang.String[] extractColNames(java.lang.String[] src,
                                                 int srcStart,
                                                 int destStart,
                                                 int length)

convertList2Array

public static int[][] convertList2Array(java.util.List<java.lang.Integer[]> list2Convert,
                                        int M,
                                        int d)

genKnotsMultiplePreds

public static double[][] genKnotsMultiplePreds(water.fvec.Frame predictVec,
                                               GAMModel.GAMParameters parms,
                                               int predIndex)

Generate knots for thin plate (TP) smoothers. Sort the first predictor column, take quatiles out of the sorted first column and grab the corresponding rows of second, third, ... predictor columns.

Parameters:

predictVec - H2OFrame containing predictor columns used to build the TP smoothers.

parms - GAMParameter

predIndex - integer denoting GAM column specificationm in parms._gam_columns

Returns:

array of knot values for predictor columns specified in parms._gam_columns[predIndex]