README.md

In yfyang86/optimise2: optim corrections.

optimise2

The optimise(stat) in R sometimes fails to converge to global optimization point. In fact, in GSL, most of these cases will report a warining message. The rule in this packages is gsl_min_fminimizer_quad_golden with more than one guess.

Dependencies: Rcpp RcppGSL

In fact, this package may only depend on libgsl. I fail to configure it on windows if Cygwin is not used.

Some c functions are modifications of survival/base.

Installation

On Linux/Mac you may simply use

library(devtools)
install_github('yfyang86/optimise2')

On windows, due to CRAN uses a strange path, you have to do the following

1. Create a folder D:\RCompile\CRANpkg\extralibs64\local (R-3.2.X uses C:\applications\extsoft)
2. Download GSL MSVC 10 and unzip it to previous path(R-3.2.X uses GSL-1.15-dev-win64)
3. Then your path include

batch D:\RCompile\CRANpkg\extralibs64\local ├───include └───lib The you can install using devtools like Mac/Linux users do.

Example

Lambda Selection

r set.seed(65535) Xmat = matrix(rnorm(100*80),ncol=80) beta0 = rnorm(80,sd=2) beta0[sample(1:80,70)] = 0. epsilon = rnorm(100) Yvec = Xmat%*%beta0 + epsilon lambdas = seq(from=.001,to=1,length.out=50) # a vec lambda_Select(Yvec,Xmat)

Example:

```r library(optimise2) N = 100 p = floor(N.8) q = p/(1:2) ratio = c(.05,.1,.2) qi=2 ri=1 ## iid case SIMU=100 RE = numeric(SIMU) for (simu in 1:SIMU){ Xmat = matrix(rnorm(Np),ncol=p) beta0 = rnorm(p,sd=2) ind.zero = sample(1:p,floor(p0.9)) ind.nzero = (1:p)[-ind.zero] beta0[ind.zero] = 0. # 90% beta <- 0 epsilon = rnorm(N) Xmat = t(t(Xmat) - apply(Xmat,2,mean)) Yvech = Xmat%%beta0 + epsilon Yvec = Yvech - mean(Yvech)

lambda_Select(Yvec,Xmat,nPsep = q[qi], perc = ratio[ri]) -> re uuz <- (abs(re$result$beta[,sum(re$selection)] )>1e-14) RE[simu] = sum((ind.nzero) %in% which(uuz))/sum(uuz) cat(simu,'-th RE: ',RE[simu],'\n') if (simu==1)plot(summary(re)[-2],type='b',col=rainbow(SIMU)[simu],pch='+',cex=.5,ylim=c(0,q[qi].8)) points(summary(re)[-2],type='b',col=rainbow(SIMU)[simu],pch='+',cex=.5,ylim=c(0,q[qi].8)) } summary(RE) #abline(h=length(ind.nzero),lty=2,col=2) ```

Test on verification set:

Linear Model with $H_0: c^\top \beta = c_0$

```r # Example 1: dim =1 x=seq( 0,10,0.1) y = matrix( 6x+7+runif( 101,min=-1,max=1) ,ncol=1) x=matrix( x,ncol=1) re = lmconst( y,x,2,19) re re[ 1] +2re[ 2]

# Example 2: multi dim x = matrix(runif(100),ncol=2) y = x %% c(6,2) +1 + rnorm( 50) re = lmconst(y=y,x=x,x0=c(2,1),y0=15) re re[1] +2re[2] +re[ 3] ```

Global Optimise: GSL routine

r f <- function(x) sin(x^2) + x/10 optimise(f, c(1, 4), tol = 1e-06) optimise(f, c(1.5, 11)) # WRONG!!! optimise2(f, c(1.5, 11), tol = 1e-06)

yfyang86/optimise2 documentation built on May 4, 2019, 2:32 p.m.