Lecture 18, 19 & 20

Question 1

Why is Correlation useful?

Answer 1

Discover relationship/possible causality

One step towards finding Causality

Question 2

What is Correlation

Answer 2

if there is a relationship between a pair/set of values.

Question 3

How can Correlations be identified visually?

Answer 3

Via Scatter Plot

Question 4

Why is Correlation important?

Answer 4

Discover Relationships

Question 5

Why is Correlation different to Causation?

Answer 5

Because data may have a similar cause such as sunglasses sales vs ice cream sales.

Question 6

What is Euclidean distance?

Answer 6

Distance between two points x and y.

Question 7

Why is Euclidean distance shit?

Answer 7

• Different scales of objects so numbers become arbitrary
• Can not discover similar behaviour at different scale
• Can not discover negative correlation

Question 8

Advantages of Pearson’s correlation

Answer 8

• Range within [ -1 , 1 ]
• Scale Invariant: r(x,y)=r(x,Ky), K is real positive constant
• Location Invariant: r(x,y)=r(x,y+C), C is real positive constant

Question 9

Disadvantage of Pearson’s correlation

Answer 9

Can not detect non linear relationships

Question 10

How is Pearson’s correlation calculated?

Answer 10

Practice that shit

Question 11

Advantages of Mutual Information?

Answer 11

• Range within [ 0, 1]

- Detect non linear relationships

Question 12

What is Variable Discretization?

Answer 12

Converting from continuous to discrete values via bins

Question 13

What methods of Variable Discretization are there?

Answer 13

Domain Knowledge, Equal-width bin, Equal frequency bin

Question 14

What is Domain Knowledge Variable Discretization?

Answer 14

Manually assigning thresholds e.g. Speed

• 0-40km/h Slow
• 40-70km/h Medium
• 70km/h+ fast

Question 15

What is Equal-width bin Variable Discretization?

Answer 15

Where bins have the same length

Question 16

What is Equal Frequency Variable Discretization?

Answer 16

Where bins have the same number of points.

Question 17

What is Entropy?

Answer 17

A measure of the Information Content

Question 18

What is Classification

Answer 18

Given a training data set find a model for classifying attributes as a function of values of other attributes.

Question 19

What is Goal of Classification?

Answer 19

To provide previously unseen data and assign a class to it.

Question 20

What is Regression?

Answer 20

Given a training data set learn a predictive model for the data.

Question 21

What is required by K Nearest neighbour Classifier?

Answer 21

• Set of records
• Metric to compute distance between records
• The value of k, i.e. the number of neighbours to retrieve.

Question 22

What is the methodology of K Nearest neighbour Classifier?

Answer 22

• Compute distance to other training records (e.g. euclidean distance/ possibly with weights)
• Identify k nearest neighbours
• Use classes of neighbours to determine the class of the unknown record

Question 23

Problems with K Nearest Neighbour?

Answer 23

• K needs to be selected carefully

- Large number of points add storage cost and search cost

Question 24

How do you calculate accuracy?

Answer 24

(TP+TN)/(TP+TN+FP+FN)

Question 25

How do decision trees work?

Answer 25

Seriously you need me to answer that ಠ_ಠ

Question 26

Problems with decision trees?

Answer 26

• Determining how to split values

- Determining when to stop splitting

Question 27

How do you specify test condition for a tree?

Answer 27

Depends on attribute types and number of splits

Question 28

How to determine the best split?

Answer 28

Nodes with homogeneous class distribution with a low level of impurity

Question 29

How is Entropy used to calculate impurity?

Answer 29

Entropy formula, 0 when all belong to one class

Question 30

How to determine how good is a split?

Answer 30

(Formula from slides) Compare impurity of parent node before split and after split