R: how to output prediction probabilities with classification models

Question

When I fit a logistic regression model with glm , I can specify type = "response" to get predicted probabilities.

model <- glm(formula= vs ~ wt + disp, data=mtcars, family=binomial)
newdata = data.frame(wt = 2.1, disp = 180)
predict(model, newdata, type="response")
        1 
0.2361081 

I'm experimenting with the logistic regression function in a new package, RSSL . Below is some sample code (from the documentation)

library(RSSL)
set.seed(1)
df <- generateSlicedCookie(1000,expected=FALSE) %>% 
  add_missinglabels_mar(Class~.,0.98)
class_lr <- LogisticRegression(Class~.,df,lambda = 0.01)
df_test <- generateSlicedCookie(1000,expected=FALSE)
predict(class_lr,df_test)

Using predict on the class_lr object gives me the class labels. And using predict(class_lr,df_test, type = "response") results in an error. Is there a way to get R to output the predicted probabilities?

Answer 1

Looking at thesource code of LogisticRegression , for predict, it calculates the prediction in log-odds ratio and converts it to probabilities and returns only the class, so there's no option for type="response" :

setMethod("predict", signature(object="LogisticRegression"), function(object, newdata) {
ModelVariables<-PreProcessingPredict(object@modelform,newdata,scaling=object@scaling,intercept=object@intercept)
  X<-ModelVariables$X

  w <- matrix(object@w, nrow=ncol(X))
  expscore <- exp(cbind(rep(0,nrow(X)), X %*% w))
  probabilities <- expscore/rowSums(expscore)

  # If we need to return classes
  classes <- factor(apply(probabilities,1,which.max),levels=1:length(object@classnames), labels=object@classnames)
  return(classes)
})

Another method associated with this class is posterior , and you can see the code is very similar, and it returns the probabilities in exp form:

setMethod("posterior", signature(object="LogisticRegression"), function(object,newdata) {

  ModelVariables<-PreProcessingPredict(modelform=object@modelform,
                                       newdata=newdata,
                                       y=NULL,
                                       scaling=object@scaling,
                                       intercept=object@intercept)

  X<-ModelVariables$X

  w <- matrix(object@w, nrow=ncol(X))
  expscore <- exp(cbind(rep(0,nrow(X)), X %*% w))
  posteriors <- expscore/rowSums(expscore)

  posteriors <- exp(posteriors)
  colnames(posteriors) <- object@classnames
  return(posteriors)
})

Apologies for the slightly long answer, if you need the probabilities, you can do:

probs = log(posterior(class_lr,df_test))

The first column is probability of being in the first class, and so on for second column. To check that the labels are similar:

pred_labels = predict(class_lr,df_test)
table(apply(probs,1,which.max) == as.numeric(pred_labels))
TRUE 
1000