1. Would a decision tree used for class probability estimation be considered a parameter learning model? Why or why not?
2. A logistic regression procedure and a support vector machine procedure were run on a set of labeled training data, resulting in the two classifiers shown in the diagram below. The purpose of the classifiers is to distinguish between two species of iris (one represented as circles and the other represented as filled circles) based on the flowers petal width and sepal width. The point with the star around it is a filled circle instance. It is starred just to highlight it, since if falls on the wrong side of the decision boundary produced by the support vector machine. Can we discern from this graph which procedure produced the better classifier? Explain the thinking behind your answer. (Hint: If your conclusion is quick and simple, you probably are not thinking about the question correctly.)

3. In the image above, the decision boundaries produced by logistic regression and the support vector machine are both linear, but they are not the same. Why are they different?
4. When optimizing the parameters for a standard linear regression model, what is the objective function (That is, what are the parameters chosen to optimize)? What is the objective function used for logistic regression (That is, what are the parameters chosen to optimize)?
5. Logistic regression outputs the log odds of a class probability. How can a probability of a class be calculated from the log odds produced by a logistic regression?

Sepal width 2 5 4 3 1 0.5 oo 0000 oo Support vector machine Petal width 0000 Oo 000000 00000 o o 1.5 Logistic regression oo Sepal width 2 5 4 3 1 0.5 oo 0000 oo Support vector machine Petal width 0000 Oo 000000 00000 o o 1.5 Logistic regression oo