How to solve a least squares (underdetermined system) quickly?

Question

I have a program in R that is computing a large amount of least squares solutions (>10,000: typically 100,000+) and, after profiling, these are the current bottlenecks for the p

Answer 1

How about Rcpp?

library(Rcpp)
cppFunction(depends='RcppArmadillo', code='
            arma::mat fRcpp (arma::mat A, arma::mat b) {
            arma::mat betahat ;
            betahat = (A.t() * A ).i() * A.t() * b ;
            return(betahat) ;
            }                                
            ')

all.equal(f1(A, b), f2(A, b), fRcpp(A, b))
#[1] TRUE
microbenchmark(f1(A, b), f2(A, b), fRcpp(A, b))
#Unit: microseconds
#        expr    min     lq     mean  median      uq     max neval
#    f1(A, b) 55.110 57.136 67.42110 59.5680 63.0120 160.873   100
#    f2(A, b) 34.444 37.685 43.86145 39.7120 41.9405 117.920   100
# fRcpp(A, b)  3.242  4.457  7.67109  8.1045  8.9150  39.307   100