Bivariate Penalized Splines for Geo-Spatial...
Transcript of Bivariate Penalized Splines for Geo-Spatial...
![Page 1: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/1.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
Bivariate Penalized Splines for Geo-SpatialModels
GuanNan WangDepartment of MathematicsCollege of William & Mary
April 20, 2016
![Page 2: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/2.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
MOTIVATION
BIVARIATE SPLINES
TRIANGULATIONS
BIVATIATE PENALIZED SPLINE ESTIMATORS
PARTIALLY LINEAR BIVARIATE SPLINE
![Page 3: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/3.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG
Damaged image with 50% of missing data observations
![Page 4: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/4.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG
Image recovered using thin-plate splines
![Page 5: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/5.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG
Face of Lena Soderberg with 8401 pixels.
![Page 6: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/6.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG
Lena Soderberg from November 1972 issue of Playboy
![Page 7: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/7.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
SPATIAL DATA AND MODELING
I Spatial is relating to the position, area, shape, and size ofthings.
I Spatial describes how objects fit together in space.I Data are facts and statistics collected together for inference
and analysis.I Spatial Data are data/information about the location and
shape of, and relationships among, geographic features,usually stored as coordinates and topology.
![Page 8: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/8.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
SPATIAL MODEL
I A common goal in spatial modeling: predicting the valueof a target variable Y over a two-dimensional domain.
I Let Xi = (X1i,X2i)ni=1 be a set of n points range over a
bounded domain Ω ⊆ R2 of an arbitrary shape.I Let Yi be the value of Y observed at point Xi.I Given n observations (Xi,Yi)n
i=1 = (X1i,X2i,Yi)ni=1, we
assume
Yi = m (Xi) + σ (Xi) εi, i = 1, · · · ,n
I εi’s are random errors and independent of Xi;I m is an unknown smooth function.
I Goal: to estimate a function of m based on the nobservations.
![Page 9: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/9.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
1-D SMOOTHING SPLINES
I The smoothing spline estimate of m is defined as asolution to the optimization problem:
n∑i=1
[Yi −m(Xi)]2 + λ
∫[m′′(t)]2dt
with λ as a fixed constant (roughness penalty parameter).
I The 1st term ensures the closeness of the estimate to thedata;
I The 2nd term penalizes the curvature of the function;
I Small λ⇒ an interpolating estimate;
I Large λ⇒ m′′(x)→ 0⇒ the least squares fit.
![Page 10: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/10.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
BIVARIATE SMOOTHING
Suppose we have two input variables X1 and X2.
I Thin-plate spline smoother: Wood (2003)I By penalizing the curvature of the spline surface, thin-plate
spline is defined as a solution to the optimization problem
n∑i=1
(Yi −m(X1i,X2i))2 + λ
∫ ∑i+j=2
(2i
)(Di
x1Dj
x2 f )2dx1dx2.
I Tensor product spline smoother:
m(x1, x2) =∑j,k
βjkBj(x1)Bk(x2).
I Useful when the data are observed on a regular grid in arectangular domain;
I Undesirable when data are located in domains withcomplex boundaries and holes.
![Page 11: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/11.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
SMOOTHING OVER DIFFICULT REGIONSRamsay (2002, JRSSB): estimate the per capita income for theIsland of Montreal, Canada.
Island of Montreal with 493 data points defined by the centroids ofcensus enumeration areas. Source: Ramsay (2002, JRSSB)
![Page 12: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/12.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
BIVARIATE SPLINES OVER TRIANGULATION
I We consider bivariate splines on triangulations to handlethe irregular domains.
Triangulation of the image of Lena Soderberg
![Page 13: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/13.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
TRIANGLE: SIZE AND SHAPE
I Let τ be a triangle, i.e., a convex hull ofthree points not located in one line.
I Given any triangle τ ,I Let |τ | be the length of its longest edge;I Let ρτ be the radius of the largest disk
inscribed in τ ;I Define the ratio βτ = |τ |/ρτ the shape
parameter of τ ;I For an equilateral triangle, βτ = 2
√3;
I Any other triangle has a larger shape ofparameter;
I When βτ is small, the triangles arerelatively uniform (all angles of trianglesin the triangulation τ are relatively thesame).
![Page 14: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/14.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
TRIANGULATIONS
A collection4 = τ1, ..., τN of triangles is called a triangulationof Ω = ∪N
i=1τi if a pair of triangles in4 intersect, then theirintersection is either a common vertex or a common edge.
Two triangulation examples.
![Page 15: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/15.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
TRIANGULATIONS
A collection4 = τ1, ..., τN of triangles is called a triangulationof Ω = ∪N
i=1τi if a pair of triangles in4 intersect, then theirintersection is either a common vertex or a common edge.
Two triangulation examples.
![Page 16: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/16.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
UNIFORM REFINEMENT OF A TRIANGULATION
Let ∆ be a given triangulation. A uniform refinement of ∆ canbe obtained by splitting each triangle τ ∈ ∆ into foursubtriangles by connecting the midpoints of the edges of τ .
A triangulation and its uniform refinement
![Page 17: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/17.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
TRIANGULATIONS IN PRACTICE
I Maxmin-angle triangulation: we seek to maximize thesmallest angle in a triangulation.
I There is no triangle that contains no data points.
I Find a polygon Ω containing all the design points of thedata and triangulate Ω by hand or computer to have atriangulation40.
I Uniformly refine40 several times to have a desiredtriangulation.
I The Delaunay algorithm is a good way to triangulate theconvex hull of an arbitrary dataset; see MATLAB program“delaunay.m”.
![Page 18: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/18.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
DEFINITION OF SPLINE FUNCTIONS
I Let τ = 〈(x1, y1), (x2, y2), (x3, y3)〉. For any pointv = (x, y) ∈ R2, let b1, b2, b3 be the solution of
x = b1x1 + b2x2 + b3x3,
y = b1y1 + b2y2 + b3y3,
1 = b1 + b2 + b3,
where coefficients (b1, b2, b3) are called the barycentriccoordinates of point v with respect to the triangle τ .
I Fix a degree d > 0. For i + j + k = d, let
Bdijk(x, y) =
d!
i!j!k!bi
1bj2bk
3 (Bernstein-Bezier polynomials).
I Let s|τ =∑
i+j+k=d cτijkBdijk(x, y), τ ∈ 4.
![Page 19: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/19.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
SPLINE FUNCTIONS ON TRIANGULATIONS
I Lai and Schumaker (2007): all basics aboutmultivariate splines
I EvaluationI DifferentiationI IntegrationI Refinement schemes of a triangulation
I Lai and Schumaker (2007): advancedproperties
I Dimension of various spline spacesI Construction of various locally supported
basis functionsI Approximation properties of various
spline spaces
Lai and Schumaker(2007, Cambridge Univ. Press)
![Page 20: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/20.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
BIVARIATE PENALIZED SPLINE ESTIMATOR
I Given λ > 0 and Xi,Yini=1, consider the minimization:
mins
n∑i=1
Yi − s (Xi)2 + λEυ(s), (1)
where
Eυ(f ) =∑τ∈4
∫τ
∑i+j=2
(2i
)(Di
x1Dj
x2 f )2dx1dx2
is the energy functional.I Let mλ be the minimizer of (1) and we call it the bivariate
penalized spline estimator over triangulation (BPSOTestimator) of m corresponding to λ.
![Page 21: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/21.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
PENALTY PARAMETER SELECTION
I Partition the original data randomly into K subsampleswith: one subsample⇒ test set, K − 1 subsamples⇒training set.
I Define the K-fold cross-validation score as
CVλ =
n∑i=1
Yi − m−k[i]
λ (Xi)2
I k[i]: the subsample containing the ith observation.I m−k[i]
λ : the estimate of the mean with the measurements ofthe k[i]th part of the data points removed.
I Select λ = arg min CVλ.
![Page 22: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/22.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG: TRIANGULATIONS
Triangulation40
![Page 23: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/23.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG: TRIANGULATIONS
Triangulation41
![Page 24: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/24.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG: TRIANGULATIONS
Triangulation42
![Page 25: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/25.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
IMAGE OF LENA SODERBERG: TRIANGULATIONS
Recovered image using bivariate splines over triangulation ∆1
![Page 26: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/26.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
MOTIVATION: CALIFORNIA HOUSE VALUE DATA
<50K50K−100K100K−150K150K−200K200K−300K>300Kboundary
I PredictionI How
differentfactors affectreal estatepropertyprices?
20,532 blocks defined by centroids of census enumeration areas (1990Census).
![Page 27: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/27.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
![Page 28: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/28.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Age
Location
![Page 29: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/29.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
![Page 30: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/30.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
Num. of Bedrooms
![Page 31: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/31.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
Num. of Bedrooms
Median Income
![Page 32: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/32.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
Num. of Bedrooms
Median Income Households
![Page 33: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/33.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
Num. of Bedrooms
Median Income Households
Population
![Page 34: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/34.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
CALIFORNIA HOUSE VALUE DATAI Data: all the block groups in California from the 1990
CensusI Target: House value
House Value
Total Num. of Rooms
Age
Num. of Bedrooms
Median Income
Location
Households
Population
![Page 35: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/35.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
6-FACTOR GLMI 6-factor GLM of house value as a linear combination of:
– House Age (Age)
– Total # of Rooms (TR)
– # of Bedrooms (BR)
– Median Income (Income)
– Population (Pop)
– Household (Hhd)
Model 1: 6-Factor GLM (Pace and Barry, 1997)
log(Value) = β0 + β1Income + β2 log(Age)
+β3 log(TR/Pop) + β4 log(BR/Pop)
+β5 log(Pop/Hhd) + β6 log(Hhd)
![Page 36: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/36.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
6-FACTOR GLM ESTIMATES
Overestimates house values in the central valley!
Underestimates the coastal enclaves of expensive houses!
Inland house values are lower than costal house values.
High values are clustered around major cities.
A. California house value data B. GLM fit with 6 factors
<50K50K−100K100K−150K150K−200K200K−300K>300Kboundary
![Page 37: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/37.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
LOCATION, LOCATION, LOCATION!I “Location matters”!I We need to adjust for the location effect:
– House Age (Age)
– Total # of Rooms (TR)
– # of Bedrooms (BR)
– Median Income (Income)
– Population (Pop)
– Household (Hhd)
– Latitude
– Longitude
Model 2: A Flexible Semiparametric Model
log(Value) = β0 + β1Income + β2 log(Age)
+β3 log(TR/Pop) + β4 log(BR/Pop)
+β5 log(Pop/Hhd) + β6 log(Hhd)
+g(Latitude,Longitude),
where g(·, ·) is a smooth bivariate function to be estimated.
![Page 38: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/38.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
ESTIMATED HOUSE VALUES
Overestimates house values in the central valley!
Underestimates the coastal enclaves of expensive houses!
Inland house values are lower than costal house values.
High values are clustered around major cities.
<50K50K-100K100K-150K150K-200K200K-300K>300Kboundary
(a) GLM fit with 6 factors (b) Bivariate penalized splines
Prediction errors of the logarithm of house values.LINEAR KRIG TPS SOAP BPST
0.146 0.083 0.081 0.079 0.052
![Page 39: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/39.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
![Page 40: Bivariate Penalized Splines for Geo-Spatial Modelsgyu.people.wm.edu/Spring2016/MATH410_Presentation.pdf · MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS](https://reader033.fdocuments.us/reader033/viewer/2022051322/60079b616e686346b521dd85/html5/thumbnails/40.jpg)
MOTIVATION BIVARIATE SPLINES TRIANGULATIONS BIVATIATE PENALIZED SPLINE ESTIMATORS PARTIALLY LINEAR BIVARIATE SPLINE
REFERENCES
I Furrer, R., Nychka, D., and Sainand, S. (2011). Package ‘fields’. Rpackage version 6.6.1. http://cran.r-project.org/web/packages/fields/fields.pdf.
I Lai, M. J. and Schumaker, L. L. (2007). Spline Functions on Triangulations.Cambridge University Press.
I Lai, M. J. and Wang, L. (2013). Bivariate penalized splines forregression. Statistica Sinica 23, 1399-1417.
I Ramsay, T. (2002). Spline smoothing over difficult regions. J R. Statist.Soc. B 64, 307-319.
I Wood, S. N. (2003). Thin plate regression splines. J R. Statist. Soc. B, 65,95-114.
I Wood, S. N., Bravington, M. V. and Hedley, S. L. (2008). Soap filmsmoothing.J R. Statist. Soc. B 70 931-955.