ecospat (version 3.3)

ecospat.cv.gbm: GBM Cross Validation

Description

K-fold and leave-one-out cross validation for GBM.

Usage

ecospat.cv.gbm (gbm.obj, data.cv, K=10, cv.lim=10, jack.knife=FALSE, verbose = FALSE)

Value

Returns a dataframe with the observations (obs) and the corresponding predictions by cross-validation or jacknife.

Arguments

gbm.obj

A calibrated GBM object with a binomial error distribution. Attention: users have to tune model input parameters according to their study!

data.cv

A dataframe object containing the calibration data set with the same names for response and predictor variables.

K

Number of folds. 10 is recommended; 5 for small data sets.

cv.lim

Minimum number of presences required to perform the K-fold cross-validation.

jack.knife

If TRUE, then the leave-one-out / jacknife cross-validation is performed instead of the 10-fold cross-validation.

verbose

Boolean indicating whether to print progress output during calculation. Default is FALSE.

Author

Christophe Randin and Antoine Guisan

Details

This function takes a calibrated GBM object with a binomial error distribution and returns predictions from a stratified 10-fold cross-validation or a leave-one-out / jack-knived cross-validation. Stratified means that the original prevalence of the presences and absences in the full dataset is conserved in each fold.

References

Randin, C.F., T. Dirnbock, S. Dullinger, N.E. Zimmermann, M. Zappa and A. Guisan. 2006. Are niche-based species distribution models transferable in space? Journal of Biogeography, 33, 1689-1703.

Pearman, P.B., C.F. Randin, O. Broennimann, P. Vittoz, W.O. van der Knaap, R. Engler, G. Le Lay, N.E. Zimmermann and A. Guisan. 2008. Prediction of plant species distributions across six millennia. Ecology Letters, 11, 357-369.

Examples

library(gbm)
data('ecospat.testData')

# data for Soldanella alpina
data.Solalp<- ecospat.testData[c("Soldanella_alpina","ddeg","mind","srad","slp","topo")] 

# gbm model for Soldanella alpina
gbm.Solalp <- gbm(Soldanella_alpina ~ ., data = data.Solalp,
                  distribution = "bernoulli", cv.folds = 10, n.cores=2)

# cross-validated predictions
gbm.pred <- ecospat.cv.gbm (gbm.obj= gbm.Solalp,data.Solalp, 
                            K=10, cv.lim=10, jack.knife=FALSE)