L10 - ACO and ML

Question 1

What is the name of the technique used when using ACO to optimise the ML models?

Answer 1

Auto-ML

Question 2

How can ACO be utilised in Machine Learning?

Answer 2

ACO can be used as a method to optimise a models parameters

ACO can be use to optimise the model itself.

Question 3

Is classification considered an optimisation problem? If so, explain how…

Answer 3

Yes, because we need to create a set of rules that optimally classifies the input data.

Question 4

What are the components of the Classification Rule? Use the rule IF(outlook=sunny) THEN playTennis = No as an example.

Answer 4

Rule Antecedent -> Condition
Rule Consequent -> Result

Question 5

A model is created by a…

Answer 5

Set of learned rules.

Question 6

Define Sequential Covering…

Answer 6

A ML technique used to build rule-based classifiers. The final set of rules make up the classification model.

Question 7

Explain how Sequential Covering works…

Answer 7

1. Start with 0 rules, and iterate over the data to derive classification rules from the training data set. The rules are accumulated, and result in a set containing all classification rules for the model.
2. This rule-based classifier can then be used on unseen data for classification tasks.

Question 8

Define the Ant Miner algorithm…

Answer 8

An ACO based data mining algorithm that classifies data instances based on the features and attributes of each instance.

Question 9

What does the Ant-Miner establish to make classifications?

Answer 9

Classification rules within the data.

Question 10

How does Ant-Miner represent its findings?

Answer 10

A graph consisting of attributes and values of the data, and the classification relationships between each of these.

Question 11

How is an Ant-Miner graph created?

Answer 11

1. Node generation:
   a) For each categorical attribute: Create an isolated node for every attribute-value in the data set.
   b) For each continuous attribute: A node is added for the attribute.
2. Ant selects a node and moves to another. If the destination node is categorical, the term is added to the rule. Otherwise, a discretisation step is performed before the term is added.

Question 12

What is the Discretisation step?

Answer 12

1. Slices the continuous attribute into finite intervals and finds the best interval for the current route.
2. Expensive due to sorting of intervals.

Question 13

What is cAnt-Miner?

Answer 13

A compressed modification of Ant-Miner algorithm that uses a simplified rule set.