What is the routine for predicting using generalized additive models including interactions with R library gam?
library("gam")
x <- data.frame(a=runif(100,1,10), b=runif(100,1,10))
x$y <- x$a*x$b
res <- gam(as.formula("y ~ s(a) + s(b)"), data=x[1:90,])
pred <- predict(res, x[91:100,], type="response")
res <- gam(as.formula("y ~ s(a) + s(b) + s(a,b,df=2)"), data=x[1:90,])
pred <- predict(res, x[91:100,], type="response")
works fine for the initial model without interactions. The latter model including interactions is learned in a meaningful way, but trying to predict results in the error:
Error in gam.s(data[["s(a, b, df = 2)"]], z, w, spar = b, df = 2, xeval = smooth.frame[["s(a, b, df = 2)"]]) : object 'b' not found