Make color ramp generation robust against two degenerate cases:
Presence of NaN values in object$frame$yval.
Duplicated values of object$frame$yval which might result in yval range in a tree being 0 which results in division by 0 in this line.
Both cases can occur in trees which were refit to new data. That is, yval values and node counts are recomputed on new data but preserving the tree structure. Unfortunately there is no built in functionality for such a "refit" in rpart, so I have my own custom code for that. Thus cannot provide a simple repro. But hopefully the PR is simple enough not to require that.
Make color ramp generation robust against two degenerate cases:
object$frame$yval
.object$frame$yval
which might result in yval range in a tree being 0 which results in division by 0 in this line.Both cases can occur in trees which were refit to new data. That is,
yval
values and node counts are recomputed on new data but preserving the tree structure. Unfortunately there is no built in functionality for such a "refit" in rpart, so I have my own custom code for that. Thus cannot provide a simple repro. But hopefully the PR is simple enough not to require that.An example of such a tree: