Incorrect number of predictions when using single record with naive bayes

[pverspeelt]

I encountered a question on SO. I reproduced the encountered error in the code below.

When you use the naive bayes method with caret and then use predict on one record, you get more predictions returned than just one. Strangely enough it only happens with the formula notation, and only with prediction type = "raw". With type = "prob" it works correctly. I also tested it with different methods, but for now it only happens with hte naive bayes method. Directly using the KlaR package works as expected.

data(iris)
library(caret)

my_control <- trainControl(method = "cv")

set.seed(42)
nbfit_default <- train(x = iris[, -5],
                       y = iris$Species,
                       method="nb",
                       trControl = my_control)


set.seed(42)
nbfit_formula <- train(Species ~ ., 
                       data = iris, 
                       method = "nb", 
                       trControl = my_control)

single_value <- (iris[2, -5])

# predict returns 4 values
predict(nbfit_formula, single_value, type = "raw")

# switch to prob and one result returns
predict(nbfit_formula, single_value, type = "prob") # result OK

# no issue with the default calls
predict(nbfit_default, single_value, type = "raw") # result OK
predict(nbfit_default, single_value, type = "prob") # result OK


 #Predict with two records or more works fine 
multiple_values <- iris[1:2,]
predict(nbfit_formula, multiple_values, type = "raw") # result OK
predict(nbfit_formula, multiple_values, type = "prob") # result OK
predict(nbfit_default, multiple_values, type = "raw") # result OK
predict(nbfit_default, multiple_values, type = "prob") # result OK

# run model direct in KlaR, no issues
set.seed(2424)
nb_formula <- NaiveBayes(Species ~ ., data = iris)
nb_default <- NaiveBayes(x = iris[, -5],
                         y = iris$Species,
                         grouping = iris$Species)

predict(nb_formula, single_value) # result OK
predict(nb_default, single_value) # result OK

[topepo]

The data were coming into the prediction function as a matrix and was getting converted to the wrong format. I've made a change to fix this.