| 
|
Readme file
SERIES
B
 Statistical
Methodology
Proportion of non-zero normal means: universal oracle equivalences and uniformly consistent estimators, by J. Jin, volume 70 (2008), pages 461–493
The code is in both R and Matlab.
Question: Given a high-dimensional vector x. For each entry of x,
the variance is 1 but the mean could be zero or non-zero. How do we
estimate the proportion of non-zero means?
Example: The file testdata.txt is an array of 100000 entries:
80% of them are samples from N(0,1); 20% them are samples from
N(u,1), where independently, |u| are sampled from Unif(1,2) and
sgn(u) are sampled from 1 and -1 with equal probabilities. The true
proportion is 0.2.
For R users:
source("Epsfunction.R.txt")
source("Epsfunctionadapt.R.txt")
x=scan("testdata.txt")
epsfunction(0.5,x,"Smooth")
epsfunctionadapt(0.015,x,"Smooth")
For Matlab users:
x = load('testdata.txt');
[t1,proprtion1] = epsfunction(0.5,x,'Smooth');
[t2,proportion2] = epsfunctionadapt(0.015,x,'Smooth');
Jiashun Jin
Department of Statistics
Baker Hall
Carnegie Mellon University
Pittsburgh
PA 15213
USA
E-mail: jiashun @stat.cmu.edu
|
Journals
SERIES
A
Statistics
in Society
SERIES
B
Statistical
Methodology
SERIES C
Applied Statistics
SERIES D
The
Statistician
|
