added an argument to limit the number of times the linesearch runs

Author: jonathan-taylor

selectiveInference/R/funs.fixed.R

@@ -3,9 +3,9 @@
 # min 1/2 || y - \beta_0 - X \beta ||_2^2 + \lambda || \beta ||_1
 
 fixedLassoInf <- function(x, y, beta, lambda, family=c("gaussian","binomial","cox"),intercept=TRUE, add.targets=NULL, status=NULL,
-sigma=NULL, alpha=0.1,
-                     type=c("partial","full"), tol.beta=1e-5, tol.kkt=0.1,
-                     gridrange=c(-100,100), bits=NULL, verbose=FALSE) {
+                          sigma=NULL, alpha=0.1,
+                          type=c("partial","full"), tol.beta=1e-5, tol.kkt=0.1,
+                          gridrange=c(-100,100), bits=NULL, verbose=FALSE, linesearch.try=5) {
 
   family = match.arg(family)
   this.call = match.call()
@@ -158,7 +158,7 @@ sigma=NULL, alpha=0.1,
 
       # Approximate inverse covariance matrix for when (n < p) from lasso_Inference.R
 
-      htheta <- InverseLinfty(hsigma, n, length(S), verbose=FALSE)
+      htheta <- InverseLinfty(hsigma, n, length(S), verbose=FALSE, max.try=linesearch.try)
       # htheta <- InverseLinfty(hsigma, n, verbose=FALSE)
 
       FS = rbind(diag(length(S)),matrix(0,pp-length(S),length(S)))
@@ -268,7 +268,7 @@ fixedLasso.poly=
 ### Functions borrowed and slightly modified from lasso_inference.R
 
 ## Approximates inverse covariance matrix theta
-InverseLinfty <- function(sigma, n, e, resol=1.2, mu=NULL, maxiter=50, threshold=1e-2, verbose = TRUE) {
+InverseLinfty <- function(sigma, n, e, resol=1.2, mu=NULL, maxiter=50, threshold=1e-2, verbose = TRUE, max.try=10) {
     isgiven <- 1;
   if (is.null(mu)){
     isgiven <- 0;
@@ -293,13 +293,12 @@ InverseLinfty <- function(sigma, n, e, resol=1.2, mu=NULL, maxiter=50, threshold
 
     output = NULL
 
-    while ((mu.stop != 1)&&(try.no<10)){
+    while ((mu.stop != 1) && (try.no<max.try) ){
       last.beta <- beta
-      #print(c("#######################trying ", try.no))
       output <- InverseLinftyOneRow(sigma, i, mu, maxiter=maxiter, soln_result=output) # uses a warm start
       beta <- output$soln
       iter <- output$iter
-      if (isgiven==1){
+      if (isgiven==1) {
         mu.stop <- 1
       }
       else{

selectiveInference/man/fixedLassoInf.Rd

@@ -10,9 +10,9 @@ fixed value of the tuning parameter lambda
 }
 \usage{
 fixedLassoInf(x, y, beta, lambda, family = c("gaussian", "binomial", 
-    "cox"),intercept=TRUE, add.targets=NULL, status=NULL, sigma=NULL, alpha=0.1,
+         "cox"),intercept=TRUE, add.targets=NULL, status=NULL, sigma=NULL, alpha=0.1,
          type=c("partial","full"), tol.beta=1e-5, tol.kkt=0.1,
-         gridrange=c(-100,100), bits=NULL, verbose=FALSE) 
+         gridrange=c(-100,100), bits=NULL, verbose=FALSE, linesearch.try=5) 
 }
 \arguments{
   \item{x}{
@@ -85,7 +85,11 @@ helpful (though computationally more costly). In particular, extra precision mig
 if the values in the output columns of \code{tailarea} differ noticeably from alpha/2.
 }
 \item{verbose}{
-Print out progress along the way? Default is FALSE}
+Print out progress along the way? Default is FALSE
+}
+\item{linesearch.try}{
+When running type="full" (i.e. debiased LASSO) how many attempts in the line search?
+}
 }
 
 \details{

selectiveInference/src/Rcpp-debias.cpp

@@ -1,6 +1,9 @@
 #include <Rcpp.h>      // need to include the main Rcpp header file 
 #include <debias.h>    // where find_one_row_void is defined
 
+// Below, the gradient should be equal to Sigma * theta + linear_func!!
+// No check is done on this.
+
 // [[Rcpp::export]]
 Rcpp::List solve_QP(Rcpp::NumericMatrix Sigma,
 		    double bound,