Unsupervised Learning

Question 1

What is clustering?

Answer 1

This is the task of grouping objects, objects in the same group are more similar.

Question 2

What are some things to consider when clustering?

Answer 2

1. The distance (similarity measure) used
2. The Number of Clusters
3. The Clustering Algorithm
4. Evaluate the Clustering Output

Question 3

What is the Single Link Measure?

Answer 3

The smallest distance between a point one cluster and a point in another cluster

Question 4

What is the Complete Link Measure?

Answer 4

The maximum distance between a point in one cluster and the a point in another cluster

Question 5

What is the Average Link Measure?

Answer 5

The average distance between points in two different clusters

Question 6

What are some issues with K-Means Clustering?

Answer 6

1. Need to specify the number of clusters in advance
2. Different initialisation of the cluster centers give different solutions
3. Unable to handle noisy data and outliers
4. Not suitable for complex data patterns

Question 7

What are the steps involved in K Means Clustering?

Answer 7

1.Compute the distance between each point and each data centre

2. Assign each point to the data centre that it is closest to
3. Move each data centre to the average of all the points that have been assigned to it
4. Repeat until convergence

Question 8

Will K Mean Clustering always converge?

Answer 8

Yes

Question 9

What is one advantage of hierarchical clustering?

Answer 9

We do not have to specify the number of groups in advance

Question 10

What are the two types of hierarchical clustering?

Answer 10

1. Agglomerative
2. Divisive

Question 11

What is Agglomerative clustering?

Answer 11

Treat each data point as a cluster

Merge the atomic clusters into larger and larger clusters

Question 12

What is Divisive Clustering?

Answer 12

Treat all data points as one single cluster

Then divide this cluster into smaller and smaller clusters

Question 13

What are the steps of Agglomerative Clustering?

Answer 13

1. Assume each point is its own cluster
2. Compute the distances (single-link, complete-link, average-link) between each cluster
3. Merge the two closest clusters
4. Repeat Steps 2 and 3 until we have one single cluster
5. Create Dendogram to show clusters at each iteration

Question 14

What is the lifetime of a cluster?

Answer 14

The distance at which the cluster was created

MINUS

The distance at which it was merged

Question 15

What are some drawbacks of Agglomerative Clustering?

Answer 15

1. Sensitive to cluster distance measure
2. Sensitive to noise
3. Less efficient

Question 16

What are some different clustering methods?

Answer 16

Centroid Based - K Means

Hierarchical - Agglomerative/ Divisive

Distribution - Assume objects from the same cluster follow the same distribution

Density Based - Define clusters as areas of higher density

Question 17

How might we test clustering models?

Answer 17

Using the within-cluster variance, and the between cluster variance

and then computing the F-Ratio

F = K *
Within_Cluster_Variance / Between_Cluster_Variance

Question 18

What is External Validation used for?

Answer 18

This is to see whether the we have actually assigned the objects to the right group