Regression subset selection, including exhaustive search.
This package performs an exhaustive search for the best subsets of a given set of potential regressors, using a branch-and-bound algorithm, and also performs searches using a number of less time-consuming techniques. It is designed to replace the "leaps()" command in S. It is based on FORTRAN77 code by Alan Miller of CSIRO Division of Mathematics & Statistics, which is described in more detail in his book "Subset Selection in Regression". Parts of the code have been published in the Applied Statistics algorithms series. There are several minor but useful improvements over the S implementation. Firstly, there is no hard-coded limit to the number of variables. Secondly, it is possible to restrict the search to subsets of at most a specified size, potentially giving a large saving in time. Thirdly, it is not necessary that the matrix of possible predictors be of full rank. This is particularly useful when there are more predictors than cases and the best "small" model is wanted. Fourthly, when there are many more cases than predictors, the search can be run on the output of biglm() and the time and memory use are then independent of the number of observations.