Instantiate a new DecisionTreeClassifier object, and use the best$\_$params$\_$ attribute of the grid$\_$search$\_$cv object to specify the best max$\_$depth, max$\_$leaf$\_$nodes and min$\_$samples$\_$split values calculated from the grid search along with a random$\_$state of $42.$ Retrain the "optimal" $($for the few parameters that we swept$)$ decision tree.

Next, use the tree.export$\_$text$()$ method to visualize the "optimal" decision tree. This function takes a trained classifier as its first parameter, and a set of feature names as its second parameter $($the feature names are included in the iris dictionary returned from the load$\_$iris$()$ function$).$ The result is a text based visualization of the decision tree. Note that this method returns a string, so you'll want to print$()$ the result to get it to look right.