Algorithm 615: the best subset of parameters in leasst absolute value regression
R. D. Armstrong, Philip O. Beck, Mabel Tam Kung · ACM Transactions on Mathematical Software · 1984
The purpose of this algorithm is to determine the best subset of parameters to fit a linear regression under a least absolute value criterion.A complete description of the algorithm is given in [2].The program consists of seven subroutines written in standard FORTRAN.During the initial phases of data analysis it is frequently desirable to consider different mathematical model formulations.One common technique in linear regression analysis is to obtain the "best" model when including exactly k independent variables.A generalization of this approach is to obtain the best subset for k = p, p + 1 . . . ., m parameters in the model, where m is the total number of independent variables observed.The solution algorithms for this best subset problem are fairly well known when the least-squares criterion is used