Jonathan

Alcantara

CS

780 Machine Learning

Assignment

0

Problem

1

The

advantages of a very flexible approach for regression or classification is

that you get a better fit for non-linear models that will decrease the

bias. For a more flexible approach

it would be preferable when we are interested in the predictors. The

disadvantages are that it requires estimating a larger number of

parameters and it increases variance.

A

less-flexible approach would be preferable when we are interested in the

inference and how well we can interpret the results.

Problem

2

1.

An application in which classification combined

with prediction may be useful is be checking if email is spam. The input would

be a sequence of common spam words and the response would indicate if an email

is spam.

2.

An application in which classification combined

with inference may be useful is figuring out which genes respond to a drug. The

input would be a gene expression and the result would be whether there is a

drug response or not.

3.

An application in which regression combined with

prediction may be useful is predicting the stock price change. The input would

be the daily news from the past years and the result would be the prediction.

4.

An application in which regression combined with

inference may be useful would be figuring out what mileage per gallon a new car

design would have. The predictors would be the how many cylinders, performance

parts and horsepower. The result would be the mileage per gallon for the car.

Problem 3

1.

Done

2.

X) Min: 1, Max: 200, Mean: 200.50

TV) Min: 0.70, Max: 296.40, Mean: 147.04

radio) Min: 0, Max: 49.600, Mean: 23.264

newspaper) Min: 0.30, Max: 144, Mean: 30.55

sales) Min: 160, Max: 27, Mean: 14.02

3.

4.