Exercise4: Hyperparameter tuning

Models in machine learning usually have multiple parameters which decide how the model looks like. A simple example is the slope and intercept of a linar model. One common task in machine learning is to find the optimal set of parameters for a particular dataset. This is called hyperparameter tuning.

source: vignette.wikia.nocookie.net/viva-deutschland

In R, the caret package provides a convinient framework for the hyperparameter tuning of various kinds of models.

Tasks:

• Train your random forest model again, but this time use caret::train(). Make sure you set the parameter importance = "impurity" (why?).
• Plot the variable importance of the model with caret::varImp.
• Get some information about the hyperparameters of random forest in a publication from Probst et. al 2019 and the help page of ranger.
• Tune one parameter of your choice by creating a caret::expand.grid() and the parameter tune.grid in caret::train().
• Plot the resulting model object. What do you see?
• Now train a model where you tune all the hyperparameters.
• Save the model as a RDS file.