Handling of case of insufficient data, following the comment of Deborah Verfaillie. The function now warns (and does no calibration) when 3 or less observation-model pairs are available for training. A new parameter na.fill has been introduced.

git log
q