David March Morla
2014-Mar-12 12:25 UTC
[R] fit and predict binomial gbm with two offset terms
Dear all, I am using the ?dismo? package to conduct boosted regression trees (BRT) for both binary and count data. The dismo package uses ?gbm? package for the implementation of BRT. I would like to incorporate two offset terms in the model, as well as being able to make predictions. For the count data I am using a Poisson model. Based on a previous post (https://stat.ethz.ch/pipermail/r-help/2010-September/253647.html), I implemented the following code: library(gbm) library(dismo) #define offset offset=(log(data$off1)+ log(data$off2)) #equivalent to log(data$off1*data$off2) #fit poisson m.pois<-gbm.step(data=data, gbm.x=7:8, gbm.y=4, offset=offset, family="poisson", tree.complexity=1, learning.rate=0.001, bag.fraction=0.7, n.folds=10) #predict poisson link <-predict.gbm(m.pois, data, n.trees=n.trees, type="link") link.offset<- link + offset pred <- exp(link.offset) My questions is how to implement the same for a binomial model? I have tried to look in different forums and documentation without success. The only clue that I have is the following document: https://r-forge.r-project.org/scm/viewvc.php/*checkout*/pkg/inst/doc/gbm.pdf?revision=18&root=gbm&pathrev=22 Any advice and/or additional references on this issue would be more than welcome. Thank you in advance, David March -- David March Morl? Spatial Ecologist Email: david at imedea.uib-csic.es IMEDEA Instituto Mediterraneo de Estudios Avanzados (UIB-CSIC) C/Miquel Marqu?s 21, 07190 Esporles, Balearic Islands. Spain www.imedea.uib.es SOCIB Balearic Islands Coastal Observing and Forecasting System Strategic Issues and Applications for Society (SIAS Division) Parc Bit, Naorte, Bloc A 2?p. pta. 3, 07121 Palma de Mallorca. Spain Tel: +034 971 43 97 64 www.socib.es SOCIAL MEDIA Google Scholar: http://scholar.google.es/citations?user=xABsDpAAAAAJ Research Gate: https://www.researchgate.net/profile/David_March3/ Linked In: http://www.linkedin.com/in/dmarch