Lecture 5

Question 1

What is the corresponding empirical risk for 0-1 loss?

Answer 1

R_0,1 (h) = 1/n En i=1 L_0,1 (y_i, h)

Question 2

If y_1, y_2, …, y_n are all unique, what is R_0,1(y_1)?

Answer 2

n-1/n

Question 3

How many parameters does does the simple linear regression model have?

Answer 3

two: w_0 and w_1

Question 4

What is a feature?

Answer 4

an attribute of data - a piece of information

Question 5

What is an example of a numerical feature?

Answer 5

maximum allowed speed, time of departure

Question 6

What is an example of a categorical feature?

Answer 6

day of the week

Question 7

What is an example of a boolean feature?

Answer 7

was there a car accident on the road?

Question 8

What is another name for H?

Answer 8

the prediction rule

Question 9

What is overfitting?

Answer 9

when we make mean squared error very small, even zero

Question 10

What is our goal when making predictions?

Answer 10

make good predictions on data we haven’t seen

Question 11

What is the syntax of quadratic regression?

Answer 11

H(x) = w_0 + w_1x_1 + w_2 x^2

Question 12

What is the syntax of exponential regression?

Answer 12

H(x) = w_0e^w_1x

Question 13

What is the syntax of constant regression?

Answer 13

H(x) = w_0

Question 14

What does w_1 stand for?

Answer 14

slope

Question 15

What does w_0 stand for?

Answer 15

intercept

Question 16

What is a big problem with R_abs(h)?

Answer 16

it is not differentiable

Question 17

How do we rewrite R_sq as a function of w_0 and w_1?

Answer 17

R_sq(w_0, w_1) = 1/n En i=1 (y_i - (w_0 + w_1x_i))^2

Question 18

What does the graph of R_sq(w_0, w_1) look like?

Answer 18

it is a loss surface