leapsOutofSampleSpatial: Linear regression model selection with an out-of-sample...

In kendonB/bellmisc: What the package does (short line)

Description

Linear regression model selection with an out-of-sample criterion for spatial extrapolation. Current version uses the minimum mean squared forecast error method. While leaps reports subsets for each number of regressors, this reports across all sizes. Randomly samples several sets of observations to be removed. Brute force algorithm without much regard for memory conservation. ***Future version should implement a parallel version of this.

Usage

leapsOutofSampleSpatial(xData, yData, nReps = 100,
  tau = ceiling(NROW(xData)/10), names = NULL, int = TRUE, nbest = 1)

Arguments

xData	a matrix of predictors
yData	a response vector
nReps	number of combinations of spatial units to be used in the estimation
tau	number of observations to remove from estimation and use in out-of-sample prediction. Default is ceiling(x/10).
int	Add an intercept to the model (default is TRUE)
nbest	number of regressions to report (default is 1)
regressor	names

Details

Thanks to Thomas Lumley for the similar leaps() function that selects with an in-sample criterion. His uses much more efficient algorithms. The current implementation is only robust for nbest=1.

Value

list of lm objects

kendonB/bellmisc documentation built on May 20, 2019, 9:03 a.m.