Statistics 526 Term 2 2002-3
Smoothing Methods in Regression

JANUARY 6
Check of class's regression background, interests
Overview of smoothing:
- density estimation and regression with ksmooth
Splus code for examples
- shapes of functions (Jarek Harezlak's thesis -
paper appears in Journal of Computations
and Graphical Statistics)
JANUARY 8
- more on shapes: John Rice's periodic astronomy data:
link to description
- families of functions:
Heckman and Zamar's cluster analysis
Biometrika
- families of functions:
Rice and Silverman's hip gait paper JRSS B
-
asymptotics for kernel density estimate
-
Homework distributed
JANUARY 13
- estimating densities of d-dimensional random vectors
- Splus example using Guy Nason's kde2D
output plot,
Splus code,
JANUARY 15
- methods for choosing the bandwidth
- plots
of density estimates of Old Faithful eruption times,
-
Splus code used in lecture
- lecture notes
- Least Squares CV plot 100 simulated
data points from standard normal
- Least Squares CV plot for the
Old Faithful data (!)
JANUARY 20
-
lecture notes
-
Splus code: Sheather-Jones and
transforming before estimating density
- Homework 2
FEBRUARY 3
- lecture notes:
boundary effect/adjustments,
higher order kernels,
-
Splus code, corrected Feb 3
- plots:
edge effect exponential data,
edge effect
exponential data
corrected boundary kernel, exponential
data,
4th order kernel, corrected Feb 3
4th order kernel, geyser data, corrected
Feb 3
FEBRUARY 10
local likelihood,
plot: local linear estimate for
geyser data,
plot: multiplication factor,
plot: difference
Splus code for HW1
FEBRUARY 12
penalized likelihood
FEBRUARY 24
Bsplines
-
lecture notes from 1998-99, Covered: sections 1 and 3.
Section 5 on smoothing splines will be covered in upcoming lectures.
- Figure comparing
bsplines and polynomials
- Splus code for the
above figure
- lecture notes
for Feb 24 and Feb 26
FEBRUARY 26
Logistic regression:
Lecture notes,
plot
Functional Data Analysis/B-spline example from biology:
-
description of problem/simulation study,
Examples
of randomly generated curves,
- difference in sample
mean curves for a simulated data set
-
permutation tests for a simulated data set:
standard approach,
function approach
- fitted regression curve
for an individual,
bspline basis used
-
Power curves:
average p-values comparing the two methods - lower curve
is from the function approach,
-
Splus code
for the simulation study
MARCH 3: Smoothing splines:
Lecture notes,
Splus code
Growth curves:
Plot 1, spline estimates
Plot 2, CV curve
logistic smoothing:
birthweight data, using gam
1998-99 notes that are of use:
Smoothing splines,
choosing smoothing parameter,
section 1 on effective number of parameters. We'll cover the rest
later.
Reproducing
Kernel Hilbert Spaces Made Easy, for those who want a
very theoretical treatment of splines (not covered in class)
MARCH 5:
Nadaraya Watson estimate, local linear estimate,
looking at hat matrix, CV and shortcut formula
lecture notes,
Splus/R code
MARCH 10:
CV (updated) (see also
Mar 3 choosing smoothing parameter 1998-99 class notes)
Asymptotics: Nadaraya-Watson,
Nadaraya-Watson class notes
1998-99
MARCH 12:
local polynomial asymptotics,
plug in choice of smoothing parameter
class notes
short
and long
versions of 1998-99 class notes
MARCH 17: gam - normal errors
class notes,
Splus hatmatrix,
Splus gam
1998-99 notes:
multivariate smoothing,
gam
MAR 19: Wei Liu,
Rice and Silverman (1991)
overheads
in your viewer: you may need to click orientation, swap landscape
to make them right-side up
MAR 24: Isabella Ghement,
partly linear models
overheads
MAR 26: Jochen Brumm, vector support machines
in microarray analysis
overheads,
plots
MAR 31 Shahadut Hossain, Bayes smoothing
overheads
APR 2 Kazi Azad, gam, logistic regression
overheads
APR 7 Kathryn Richardson, exponential weighted averages
and kernel smoothing
overheads
APR 9 Mushfiqur Rahman, heteroscedasticity in
regression
overheads,
plots
APR 11, 11-2 Lindsey Turner,
SiZer
overheads,
and Mike Danilov
longitudinal analysis
overheads,
Pizza
|