how to get probabilities between 0 and 1 using glmnet categorical regression

Question

My response variable is categorical, from 1 to 7. I understand using glmnet I can set the type to response and get the probability of the prediction.

prob.vec = predict.cv.glmnet(cvfit, newx = X.test, s = "lambda.min", type = "response")

However, I am interested in having the probability of other category as well. I wonder if such functionality exist in glmnet.

Answer 1

Set family = 'multinomial' within cv.glmnet and you should be all set. Here is a simple example.

library(glmnet)
y <- iris$Species
x <- as.matrix(iris[,1:4])
m1 <- cv.glmnet(x, y, family = 'multinomial')

predict(m1, newx = x, s = 'lambda.min', type = 'response')[1:2,,]
#        setosa  versicolor    virginica
#[1,] 0.9992059 0.000794057 2.548623e-20
#[2,] 0.9961627 0.003837330 1.206001e-18