Note that y is correlated with both x_1 and x_2 but the two predictors are independent of each other, as seen by cor(dat). Set the seed to 1, then use the caret package to partition into a test and training set of equal size. Compare the RMSE when using just x_1, just x_2 and both x_1 and x_2. Train a single linear model for each (not 100 like in the previous questions). Which of the three models performs the best (has the lowest RMSE)