CVgam function

Cross-validation estimate of accuracy from GAM model fit

The cross-validation estimate of accuracy is sufficiently independent of the available model fitting criteria (including Generalized Cross-validation) that it provides a useful check on the extent of downward bias in the estimated standard error of residual.

CVgam(formula, data, nfold = 10, debug.level = 0, method = "GCV.Cp",
              printit = TRUE, cvparts = NULL, gamma = 1, seed = 29)

Arguments

• formula: Model formula, for passing to the gam() function
• data: data frame that supplies the data
• nfold: Number of cross-validation folds
• debug.level: See gam for details
• method: Fit method for GAM model. See gam for details
• printit: Should summary information be printed?
• cvparts: Use, if required, to specify the precise folds used for the cross-validation. The comparison between different models will be more accurate if the same folds are used.
• gamma: See gam for details.
• seed: Set seed, if required, so that results are exactly reproducible

Returns

• fitted: fitted values

• resid: residuals

• cvscale: scale parameter from cross-validation

• scale.gam: scale parameter from function gam

The scale parameter from cross-validation is the error mean square)

References

https://maths-people.anu.edu.au/~johnm/nzsr/taws.html

Author(s)

John Maindonald

Examples

if(require(sp)){
library(mgcv)
data(meuse)
meuse$ffreq <- factor(meuse$ffreq)
CVgam(formula=log(zinc)~s(elev) + s(dist) + ffreq + soil,
              data = meuse, nfold = 10, debug.level = 0, method = "GCV.Cp",
              printit = TRUE, cvparts = NULL, gamma = 1, seed = 29)
}