Class 6 Practice Questions BLUM with Answers

pdf

School

University of Toronto *

*We aren’t endorsed by this school

Course

1215

Subject

Statistics

Date

Apr 3, 2024

Type

pdf

Pages

Uploaded by ChefMorning59

Class 6 Practice Questions 1. Consider the scatterplot below. If the ML prediction model indicated by the linear regression line is used (select all that apply), a. The ML model suffers from bias (underfitting) at x=0 b. The ML model is likely to suffer from variance (overfitting) at x=7 No, because the model is indicated by the regression line. That line extrapolates out to x=7, fit mostly from data where x<7. c. The feature x has no predictive power for outcome y d. Using a series of dummy variables, one for each value of x, will eliminate the bias of this ML model No, because no matter what prediction is made for each x, there is still a lot of dispersion in y for each x. 2. You have fit an ML regression model that you find suffers from a high bias (underfitting). To help reduce this bias, your best option of those listed below is to a. Add features that have correlations near 1 or -1 with the features already in the model b. Add features that have correlations near 0 with the features already in the model c. Collect additional data on the features you already have in the model to add observations 0 1 2 3 4 5 y 0 2 4 6 8 10 x

3. You have fit an ML regression model that you find suffers from a high variance (overfitting). To help reduce this variance, your best option of those listed below is to a. Add features that have correlations near 1 or -1 with the features already in the model b. Add features that have correlations near 0 with the features already in the model c. Collect additional data on the features you already have in the model to add observations 4. You are working as an ML analyst for the 7-Eleven chain of convenience stores. 7-Eleven is in a few provinces, including Ontario and Saskatchewan, and is trying to predict store level sales if it were to enter the Manitoba market. You have come up with a model that fits well in the training data set (has low bias) when trained using data from Ontario only. You also have data for Saskatchewan stores but have not used it. (Suggested answers below are not the only possible answers.) a. What is the potential difficulty with using your prediction model trained on Ontario data to make predictions about store sales in Manitoba? This is an issue of external validity. There may be unobserved differences in attributes between Manitoba and Ontario that would generate differences between predictions and actual outcomes in Manitoba. The two contexts are different. For example, in Manitoba (even Winnipeg), customers would be more likely to drive to the 7-Eleven than in Ontario (Toronto). The customer base might also have interests in purchasing different types of products in the two provinces. Training the model on Ontario only could miss these things. b. With the model trained on Ontario, is there something else you can do to test it? Yes, you could see how well it predicts for the Saskatchewan stores. As Saskatchewan is probably more similar to Manitoba than Ontario, a good fit for Saskatchewan may raise confidence that the model would do well for Manitoba as well. Another thing you could do is to retrain the same model on the Saskatchewan data and see if it has a set of predictions that are different. This is analogous to the Dufferin Grove – Forest Hill South comparison in class. 5. You observe the following counts of high spending - non high spending customers by gender. Male Female High Spending 5 1 Non high spending 5 9 What is Pr(high spending|male)? 0.5 What is Pr(high spending|female)?

0.1 Are High Spending and Gender independent? No. If they were, Pr(male)Pr(high spending)=0.5*0.1 = 0.05 would equal Pr(male,high spending) = 0.25. You can also see that high spending is correlated with gender, as males are more likely to be high spending. 6. You are doing research on employment practices at your competitor firm, GelCo. You have been tracking some of GelCo’s new hires over the past few years and determine the probability of an employee leaving GelCo conditional on being at the firm for less than 1 year is 0.30. You also notice in GelCo’s annual report a statement that 50% of GelCo employees have been at the firm for less than one year. What fraction of GelCo's total employees do you predict have been at the firm for <1 year and will leave within 1 year of being hired? You can just calculate the joint probability of beign at the firm for less than one year and leaving. (use Bayes’ Rule if you prefer). Pr(leave|<1yr)=0.3 = Pr(leave,<1yr)/Pr(<1yr) = ?/0.5. So Pr(leave,<1yr) = 0.15. Alternatively, you can think intuitively that since 30% of the new employees leave and 50% of employees are new, out of all employees 50%*30% = 15% or 0.15 are both new and will leave within 1 year. 7. Why do we typically prefer to use logistic regression over regression for classification? (not required yet) Because regression can yield predicted class probabilities that are <0 or >1. Logistic regression avoids this problem.

Your preview ends here

Eager to read complete document? Join bartleby learn and gain access to the full version

Access to all documents
Unlimited textbook solutions
24/7 expert homework help

Related Documents

part 2 project math 309Cs.docx

Multiple_Choice_9.1-9.3_no_answers-2.rtf

phys 223 lab 1 report.docx

Course Hero PSC 402 Chapter 12 .pdf

Central Tendency, Mean, Median, Standard Deviation.docx

Descriptive Statistics.docx

STATS 10 Assignment 2 (1).pdf

gamble_-_iris_ebp_parts_3_and_4_-_11-26.pdf

CIV171-SPRING-2024-ASSIGNMENT-2-GONZALO-LAVALLE.PDF

STAT100 HW#3.pdf

Practice Week 10.pdf

Quiz 10.pdf

Recommended textbooks for you

College Algebra

Algebra

ISBN:9781305115545

Author:James Stewart, Lothar Redlin, Saleem Watson

Publisher:Cengage Learning

Algebra and Trigonometry (MindTap Course List)

Algebra

ISBN:9781305071742

Author:James Stewart, Lothar Redlin, Saleem Watson

Publisher:Cengage Learning

Linear Algebra: A Modern Introduction

Algebra

ISBN:9781285463247

Author:David Poole

Publisher:Cengage Learning

College Algebra

Algebra

ISBN:9781337282291

Author:Ron Larson

Publisher:Cengage Learning

Glencoe Algebra 1, Student Edition, 9780079039897...

Algebra

ISBN:9780079039897

Author:Carter

Publisher:McGraw Hill

Functions and Change: A Modeling Approach to Coll...

Algebra

ISBN:9781337111348

Author:Bruce Crauder, Benny Evans, Alan Noell

Publisher:Cengage Learning

SEE MORE TEXTBOOKS

Recommended textbooks for you

College Algebra
Algebra
ISBN:9781305115545
Author:James Stewart, Lothar Redlin, Saleem Watson
Publisher:Cengage Learning
Algebra and Trigonometry (MindTap Course List)
Algebra
ISBN:9781305071742
Author:James Stewart, Lothar Redlin, Saleem Watson
Publisher:Cengage Learning
Linear Algebra: A Modern Introduction
Algebra
ISBN:9781285463247
Author:David Poole
Publisher:Cengage Learning
College Algebra
Algebra
ISBN:9781337282291
Author:Ron Larson
Publisher:Cengage Learning
Glencoe Algebra 1, Student Edition, 9780079039897...
Algebra
ISBN:9780079039897
Author:Carter
Publisher:McGraw Hill
Functions and Change: A Modeling Approach to Coll...
Algebra
ISBN:9781337111348
Author:Bruce Crauder, Benny Evans, Alan Noell
Publisher:Cengage Learning