Question 1

Consider the following two charts (chart 1 and chart 2) that were generated from two di erent models and datasets. What do these charts tell you? Was the underlying model likely regression or classification for each? How might you numerically measure each of the two models (would you use RMSE or accuracy)?

Question 2

How might you encode each of the following values, what function would you use? Would it be di erent if you were trying to use the value as input (x) or as the prediction (y).

• Age
• Favorite Color
• Length
• Gender

What problems might you encounter encoding a value like zip code (as input)? How would you address these problems?

Midterm Page 1

Question 3

How would your decision to use validation/holdout or k-fold cross validation a ect your ability to obtain out-of-sample predictions from the neural network? Why are out of sample predictions important? What is overfitting, and how do you prevent it?

Question 4

What is the e ect of using too large of a learning rate? What is the e ect of too small? What role does momentum play in backpropagation? How does ADAM relate to learning rate and momentum?

Question 5

Give an example of a classification problem that you might use a neural network for. Give an example of a regression problem you might use a neural network for. For both, describe how you would evaluate the neural network.