Title: | Adaptive SURE Thresholding Using Side Information |
---|---|
Description: | Provides the ASUS procedure for estimating a high dimensional sparse parameter in the presence of auxiliary data that encode side information on sparsity. It is a robust data combination procedure in the sense that even when pooling non-informative auxiliary data ASUS would be at least as efficient as competing soft thresholding based methods that do not use auxiliary data. For more information, please see the paper Adaptive Sparse Estimation with Side Information by Banerjee, Mukherjee and Sun (JASA 2020). |
Authors: | Trambak Banerjee [aut, cre], Gourab Mukherjee [aut], Wenguang Sun [aut] |
Maintainer: | Trambak Banerjee <[email protected]> |
License: | GPL(>=2) |
Version: | 1.5.0 |
Built: | 2025-02-15 04:24:25 UTC |
Source: | https://github.com/trambakbanerjee/asus |
ASUS procedure for shrinkage estimation of a high dimensional sparse parameter.
asus(d, v.d, s, k = 2, m = 50)
asus(d, v.d, s, k = 2, m = 50)
d |
an n vector of primary observations |
v.d |
an n vector of variances for each component of d |
s |
an n vector of side information |
k |
number of groups. Default is k=2 |
m |
partitions the support of |
Estimates a sparse high dimensional vector using the ASUS procedure described in Banerjee et al. (2017).
If k = 1 then ASUS is the SureShrink estimator. The current implementation of ASUS estimates the grouping thresholds
based on the magnitude of . See the reference for more details.
est - an n vector holding the estimates
mse - estimate of risk
tau - k-1 vector of grouping parameters if k>=2
t - k vector of thresholding parameters
size - k vector of group sizes
Banerjee. T, Mukherjee. G and Sun. W. Adaptive Sparse Estimation with Side Information. Journal of the American Statistical Association 115, no. 532 (2020): 2053-2067.
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) set.seed(42) s<-rnorm(10,3,0.1) asus.out<-asus(d,v.d,s)
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) set.seed(42) s<-rnorm(10,3,0.1) asus.out<-asus(d,v.d,s)
Estimates the risk of asus when there are k(>2) groups with pre-defined grouping thresholds
asus.cuts(d, v.d, s, cutpoints)
asus.cuts(d, v.d, s, cutpoints)
d |
an n vector of primary observations |
v.d |
an n vector of variances for each component of d |
s |
an n vector of side information |
cutpoints |
k-1 pre-defined grouping thresholds for k groups. k must be bigger than 2. |
Estimates the risk of asus when there are k(>2) groups
with k pre-defined grouping thresholds. This function is called when asus
executes.
mse - estimate of risk
Banerjee. T, Mukherjee. G and Sun. W. Adaptive Sparse Estimation with Side Information. Journal of the American Statistical Association 115, no. 532 (2020): 2053-2067.
asus
,sureshrink
,ejs
,sureshrink.mse
library(asus) set.seed(42) d<-rnorm(10) v.d<- rep(1,10) set.seed(42) s<-rnorm(10) out<-asus.cuts(d,v.d,s,c(0.1,0.5,1))
library(asus) set.seed(42) d<-rnorm(10) v.d<- rep(1,10) set.seed(42) s<-rnorm(10) out<-asus.cuts(d,v.d,s,c(0.1,0.5,1))
Extended James-Stein estimator of a high dimensional sparse parameter.
ejs(d, v.d)
ejs(d, v.d)
d |
an n vector of observations |
v.d |
an n vector of variances for each component of d |
Extended James-Stein estimator of mean from Brown (2008) and equation (7.3) in Xie et al. (2012)
est - an n vector holding the estimates
Brown, L.D. (2008). In-Season Prediction of Batting Averages: A Field Test of Empirical Bayes and Bayes Methodologies. The Annals of Applied Statistics, 2, 113-152
Xie, X. C., Kou, S. C., and Brown, L. D. (2012). SURE Estimates for a Heteroscedastic Hierarchical Model. Journal of the American Statistical Association, 107, 1465-1479.
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) theta.hat<-ejs(d,v.d)
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) theta.hat<-ejs(d,v.d)
Soft thresholds the input signal y with the threshold value thld
softTh(y, thld)
softTh(y, thld)
y |
1D signal to be thresholded |
thld |
numeric threshold value |
a numeric vector of thresholded values of the same length as y.
Donoho, David L. "De-noising by soft-thresholding." IEEE transactions on information theory 41, no. 3 (1995): 613-627.
library(asus) set.seed(42) y<-rnorm(10,2,1) thld<- 3 x<-softTh(y,thld)
library(asus) set.seed(42) y<-rnorm(10,2,1) thld<- 3 x<-softTh(y,thld)
SureShrink estimator of a high dimensional sparse parameter from Donoho and Johnstone (1995)
sureshrink(d, v.d)
sureshrink(d, v.d)
d |
an n vector of observations |
v.d |
an n vector of variances for each component of d |
Estimates a threshold t by minimizing the SURE function and then soft thresholds d using t.
est - an n vector holding the estimates
t - estimated threshold
David L Donoho and Iain M Johnstone. Adapting to unknown smoothness via wavelet shrinkage. Journal of the american statistical association, 90(432):1200-1224, 1995
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) theta.hat<-sureshrink(d,v.d)
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) theta.hat<-sureshrink(d,v.d)
Stein's Unbiased Risk Estimate for the sureshrink estimator
sureshrink.mse(d, v.d, type = 1, t = 0)
sureshrink.mse(d, v.d, type = 1, t = 0)
d |
an n vector of observations |
v.d |
an n vector of variances for each component of d |
type |
set type=1 if you want the thresholding parameter t to be estimated. Otherwise set type = 0 in which case you must provide t. Default is type = 1 |
t |
soft thresholding parameter. If type = 1, then t is estimated whereas if type = 0 then you must provide t. Default is t = 0 (and type = 1) |
Estimates the risk of the surehsrink estimator of Donoho and Johnstone (1995).
sure.est - SURE estimate of risk
t - estimated threshold (meaningless if type = 0)
Charles M Stein. Estimation of the mean of a multivariate normal distribution. The annals of Statistics, pages 1135-1151, 1981
David L Donoho and Iain M Johnstone. Adapting to unknown smoothness via wavelet shrinkage. Journal of the american statistical association, 90(432):1200-1224, 1995
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) mse<-sureshrink.mse(d,v.d)
library(asus) set.seed(42) d<-rnorm(10,2,1) v.d<- rep(1,10) mse<-sureshrink.mse(d,v.d)