[R-sig-ME] unable to predict from GAM model after warning on complete separation

[David Villegas Ríos]

Dear list.
I´m fitting some logistic bam models, and for some of them I get the
following warning:

Warning message:
In bgam.fitd(G, mf, gp, scale, nobs.extra = 0, rho = rho, coef = coef,  :
  fitted probabilities numerically 0 or 1 occurred

However the model seems to converge and the summary(), plot() and gam.check
() functions produce apparently the right results.

My objective is to make some predictions using those models, but when I get
the warning above, the predict.bam is not working and yields the following
error:

Error in Predict.matrix.tprs.smooth(object, dk$data) :
  no data to predict at

Is there any workaround to solve this issue and get the predictions, or
predictions are simply not possible from models where probabilities
numerically 0 or 1 occur? I have no problem predicting from models that do
not produce the warning.

Thanks,

David

	[[alternative HTML version deleted]]