## Modeling and evaluation

With the data prepared, we will begin the modeling process. For comparison purposes, we will create a model using best subsets regression like the previous two chapters and then utilize the regularization techniques.

### Best subsets

The following code is, for the most part, a rehash of what we developed in Chapter 2, *Linear Regression - The Blocking and Tackling of Machine Learning*. We will create the best subset object using the `regsubsets()`

command and specify the `train`

portion of `data`

. The variables that are selected will then be used in a model on the `test`

set, which we will evaluate with a mean squared error calculation.

The model that we are building is written out as `lpsa ~ .`

with the tilde and period stating that we want to use all the remaining variables in our data frame, with the exception of the response:

**> subfit <- regsubsets(lpsa ~ ., data = train)**

With the model built, you can produce the best subset with two lines of code. The first one turns the `summary...`