zjph602xtc / wsvm Goto Github PK

When using the predict method on a wsvm object with newdata that has NA values in some columns the call produces an error. The problem happens when assigning names to ret2:

m = WeightSVM::wsvm(cyl~., weight=1:nrow(mtcars), data=mtcars)
mtcars2 = mtcars
mtcars2$mpg[1] = NA
predict(m, newdata=mtcars2)

Error in names(ret2) <- rowns : 
  'names' attribute [32] must be the same length as the vector [31]

Hi,

I've been playing around the WeightSVM package and noticed something odd.

I trained a model, on a three-class dataset, using the option probability = TRUE

I then tried to predict on a new data set.

In the predict function,

• when specifying probability = FALSE, I get a class prediction for every row.
• when specifying probability = TRUE, the class predictions are much different than the previous one.

Is this normal behavior?
It looks like the cutoff is different in the two methods?