error when predict probabilities on model trained on a dgCMatrix

[anmiko]

when a model is trained on a dgCMatrix (e.g. created by hashed.model.matrix of FeatureHashing )

an attempt to call predict with type ="prob" raises error

Error in as.data.frame.default(unkX) :
cannot coerce class "structure("dgCMatrix", package = "Matrix")" to a data.frame

but predict works fine when type='raw". Also as a workaround probFunction can be called directly.

code example

library(glmnet)
library(FeatureHashing)
library(caret)
tmp.df <- data.frame(A=1:12, B=3:14, Y = c('a','a','b','c','b','a','a','a','c','c','b','a'))

mat.fh <- hashed.model.matrix(~ A+B, tmp.df, 2^10, create.mapping = TRUE)

model <- train(x = mat.fh, y = as.factor(tmp.df$Y), method = 'glmnet')

probs <- predict(model, newdata = mat.fh, type = "prob")

#a workaround  
probs <- probFunction(model$modelInfo, model$finalModel, mat.fh, model$preProcess)

caret version '6.0.62'

[topepo]

Thanks. This is a bug and has been fixed.