STATS 415 001 FA 2021 Quizzes Mock midterm!

Mock midterm

Started: Oct 13 at 9:11pm

Quiz Instruc!ons

Quiz saved at 9:11pm

This exam has 5 T/F ques!ons, 5 single-choice ques!ons, 3 mul!ple-choice ques!ons and 2 data analysis ques!ons. The final

two data analysis ques!ons require submission of rmd files that answer all the ques!ons.

2 ptsQues!on 1

True

False

The larger K you choose in the K-nearest-neighbors (KNN) method, the more accurate the resul!ng classifier

will become.

2 ptsQues!on 2

True

False

A credit card company would like to hire you to build an automa!c system that tells whether a given

applicant should be issued with a credit card. This task is an unsupervised learning problem.

2 ptsQues!on 3

True

False

The more covariates you recruit in your linear regression model, the less training error you will get.

2 ptsQues!on 4

True

False

The sample median of solves a least square problem in the sense that () { } ∈[]

() = ∈(−∞, ∞) ∑=1 ( − )

2

2 ptsQues!on 5

True

False

The sample mean is sensi!ve to outliers.

3 ptsQues!on 6

The ordinary least squares (OLS) es!mator does not have any bias.

The higher the on the training data is, the stronger generaliza!on capabitlity of the learned model has.

The residuals are independent across the observa!ons.

The residuals are uncorrelated across the observa!ons.

Which of the following is true about linear regression?

2

3 ptsQues!on 7

It will not change

It will increase by 1

It will increase by some amount, but we do not have enough informa!on to determine by how much

Suppose I have a sample of 100 data points, all of them unique posi!ve numbers. If I mul!ply 10 largest

values by 100, how will the sample mean change?

3 ptsQues!on 8

It is a discrimina!ve method rather than a genera!ve one in the sense that it does not involve the data genera!on

mechanism.

It can be applied to prac!cal data that are not exactly Gaussian.

It is mo!vated by the Bayes Rule.

Compared with general QDA, LDA requires fewer parameters to be es!mated.

Which of the following is false about the linear discriminant analysis (LDA)?

3 ptsQues!on 9

The smaller K is, the lower the training MSE

The larger K is, the more robust the resul!ng classifier is.

The larger K is, the lower the tes!ng error is.

The smaller K is, the more sensi!ve the resul!ng classifier is.

Which of the following is false about KNN?

3 ptsQues!on 10

A sca"er plot

Side-by-side boxplots of Y for each level of X

Side-by-side boxplots of X for different ranges of Y

Suppose you have two variables, X and Y, and are interested in visualizing their rela!onship. X is categorical

with three levels, and Y is a con!nuous variable that takes values between 0 and 1. What is the most

appropriate plot to make?

5 ptsQues!on 11

In a two-class classifica!on task, the feature vectors from the two classes are Gaussian and share the same covariance

structure.

Method (A): LDA v.s. Model (B): QDA.

In a two-class classifica!on task, the feature vectors from the two classes are Gaussian and have different means.

Method (A): LDA v.s. Model (B): QDA.

In a classifica!on problem, the features are independent of each other.

Method (A): QDA v.s. Method (B): Naive Bayes

In a regression problem with a con!nuous response Y and a single covariate , the sca"er plot of and is presented

as follows:

Method (A): KNN v.s. Method (B): Linear regression

In each of the following setups, !ck the box if Method (A) is preferred to Method (B).

5 ptsQues!on 12

(M3) is problema!c because we need to introduce extra dummy variables to express , which is discrete.

In (M1), must be non-nega!ve.

According to (M2), one unit increase in , given all the other predictors fixed, implies increase in . In words, one

more USD expense on produc!on will improve the ra!ng by on average.

The in-sample of (M3) will be smaller than that of (M1) and (M2).

Suppose you are trying to predict the average ra!ng ( , a con!nuous response variable ranging from 0 to 10)

of a movie based on whether the movie is a roman!c movie ( : 1 if yes and 0 otherwise), its produc!on cost

( : con!nuous, in USD) and the number of views ( : a discrete non-nega!ve variable). Consider the

following three models under the fixed design setup:

(M1):

(M2):

(M3):

where . Which of the following arguments is true?

1

2 3

= + + ,0 1 1

= + + + + ,0 11 2 2 312

= + + + + + ,0 11 2 2 33 412

() = 0

3

0

2 2

2

2

5 ptsQues!on 13

The projec!on matrix is symmetric.

The projec!on matrix is always full-rank and thus inver!ble.

Given a design matrix , whose rows correspond to the feature vectors of observa!ons, and a

response vector , which of the following arguments is true?

∈ ℝ ×

∈ ℝ

= ( ∈⊤ )−1⊤ ℝ×

( − = − )2

( − = 0)100 100

( − ) ≥ 0⊤

≥ 0⊤

30 ptsQues!on 14

Upload

Our goal in this problem is to predict the housing price using its different features. Download and read

"housing.csv" in the folder "Mock midterm datasets" using R. Submit your R markdown file to answer the

following ques!ons.

(1) Choose the first 10,000 observa!ons of the loaded data frame as the training data and the rest as the

tes!ng data. Build a linear model on the training data by regressing the housing price on these variables:

bedrooms, bathrooms, sq$_living and sq$_lot, including an intercept term. Report the coefficients you

obtained. What are the training and tes!ng MSEs?

(2) Add the zipcode to your linear model. Report the training and tes!ng MSEs of your new model. How many

parameters are there in this new model? (Hint: Pay a"en!on to the data type of the zipcode.)

Choose a File

30 ptsQues!on 15

Upload

Download and read "iris.csv" in the folder "Mock midterm datasets" using R. Submit a R markdown file that

answers the following ques!ons.

(1) Give the sca"er plot of the features "Petal.Width" and "Petal.Length" of all the observa!ons. Use three

different colors to represent the three species, and describe correpondence between the colors and species

in the legend of the plot.

(2) Choose the first 100 observa!ons of the loaded data frame as the training data and the rest as the tes!ng

data. Apply LDA and QDA using only the features "Petal.Width" and "Petal.Length". Report the training and

tes!ng predic!on accuracy of the LDA and QDA methods respec!vely.

(3) Apply the K-NN method on the same training and tes!ng dataset as in (2). Give the plot of the tes!ng

MSE with respect to choice of K.

Choose a File

Submit Quiz

Ques!ons

Time Running: Hide

1 Hour, 19 Minutes, 42 Seconds

? Ques!on 1

? Ques!on 2

? Ques!on 3

? Ques!on 4

? Ques!on 5

? Ques!on 6

? Ques!on 7

? Ques!on 8

? Ques!on 9

Fall 2021

Home

Announcements

Assignments

Grades

People

Pages

Quizzes

Modules

Piazza

Account

Dashboard

Courses

Groups

Calendar

Inbox

History

Help

Well-being

14

欢迎咨询51作业君