WEEK Five

Question 1

Classifying a new case in BN

Answer 1

Take value for all of the attributes and introduce it as evidence- clamp the corresponding BN node to that value

Question 2

Goal of a classifier

Answer 2

To predict an outcome class based on a set of attributes

Question 3

Describe Machine Learning (ML)

Answer 3

Learning from experience with respect to task and performance measures

Question 4

Model Training Process

Answer 4

Modifies overtime as you learn something- trying to build a better representation of the data
It is going to classify something for us

Question 5

Model Testing Process

Answer 5

Makes a prediction without seeing the actual answer.
The human has access to the answer to know if the model is right or not.
Ex. Self driving cars.

Question 6

Model Application

Answer 6

Model is used on data where the answer is not known

First we train, then we test, then we model application

Question 7

What data to use for training vs testing- Holdout Method

Answer 7

Split the data into a training and testing set

Typically use more data for training than testing - but various splits possible

Question 8

What data to use for training vs testing- Cross Validation Method

Answer 8

Equal size parts (folds)

We train and test on every portion of the data then calculate the average of all 4.

Question 9

Calculating Accuracy & The Limitations

Answer 9

# of correctly classified cases/ total number of cases 
Does not tell us how good the model is at predicting a certain class just tells us about the overall model of performance. 

Can be correct: When it predicts __ and is is __
Can be incorrect: When it predicts __ but it is actually ___

Question 10

Confusion Matrix

Answer 10

How good is the model at making predictions 
Fine grain information on a performance for the class under consideration. 

P REFERS TO THE MODEL UNDER PREDICTION 
N MEANS THE MODEL DIDN'T PREDICT IT 
TP - Predicts robin, it is robin 
TN- Predicts not robin, it is not robin 
FP- Predicts robin, it is a raven 
FN- Predicts not a robin, is a robin

Question 11

Baseline Model

Answer 11

Predicts the most frequent outcome without taking into account any attributes- predicts the more frequent outcome

Question 12

True positive rate (TPR)

Answer 12

True Positives/ True positives + false negatives

Good to have a high true positive rate- its predicting what it is supposed to predict. False Negatives have to be low in order to have a high TPR

Note * TPR is not TP

Question 13

True Negative Rate (TNR)

Answer 13

True negatives / True negatives + false positives