1. After fitting a model, we evaluate it on holdout data in order to

• A. increase generalization performance
• B. decrease the model's complexity
• C. estimate generalization performance
• D. assess the model's complexity

2. When we change (you can think of this as "dialing up" or "dialing down") the classification threshold for a binary classifier, which of the following happens?

• A. It affects the classification of some instances from False Positive to False Negative (and vice versa).
• B. It affects the classification of some instances from True Positive to True Negative (and vice versa).
• C. It affects the classification of some instances from False Negatives to True Positives (and vice versa).
• D. All of the above

3. We are about to deploy a classification system on a dataset. Obviously we don't know the true class membership of any of the instances in that dataset, but we do know that in this dataset exactly 50% of the instances are positive and the other 50% are negative. We classify the examples using a binary classifier that is 80% accurate for both the positive and negative classes. The lift of the classifier is:

• A. 4
• B. 1.8
• C. 1.6
• D. None of the above