I have the product of cross-validated regression, with the object being of class train; it used 'glmnet' as the method.
objModel <- train(trainDF[,predictorsNames], trainDF[,outcomeName],
method='glmnet', metric = "RMSE", trControl=objControl)
plot(varImp(objModel,scale=F))
When I plot the variable importance using VarImp, it accurately shows the importance of the variables, but it indicates them all in the positive direction. For instance, I know that 'lib' and 'cohort:Millenial' are negative predictors, but of high magnitude.
Every walkthrough and aid I have seen/used automatically plots the variables importance in the proper direction with plot(VarImp), with negative importance variables going left on the x-axis from the origin and positive importance variables going right from the origin; however, all of the variables are plotted as positive importance. Proper magnitude, but wrong direction.
Is there an in-depth fix for this problem? The documentation for VarImp remarks that glmnet objects get assigned importance based on the absolute value of the t-statistic, but that doesn't explain why I see the code run properly in depicting negative importance elsewhere. Thanks so much.