Unsupervised Learning

Question 1

What is unsupervised learning?

Answer 1

Learning relationships in the data without having any ground truth to evaluate it with

Question 2

Why would you want to use an unsupervised learning algorithm?

Answer 2

Segmentation, criminal activity identification, identifying new species, creating classes for a classification algorithm

Question 3

What is clustering?

Answer 3

Grouping points together based on some distance function

Question 4

What is community detection?

Answer 4

Taking a graph of nodes and determining communities of objects using metrics like similarity, distance, eigenvectors etc.

Question 5

What is topic modelling?

Answer 5

A technique of detecting topics within data, by grouping words together

Question 6

What is an example of an unsupervised learning algorithm? (pick one)

Answer 6

K-means clustering, DBSCAN, hierarchical clustering, hard/soft clustering

Question 7

How does DBSCAN cluster objects together?

Answer 7

Finding core regions of high density, and expanding clusters from them

Question 8

How does hierarchical clustering cluster objects together?

Answer 8

Splitting clusters iteratively into two groups until we have groups of classes

Question 9

What does it mean for a clustering algorithm to be hard?

Answer 9

Each object belongs in one cluster only

Question 10

What does it mean for a clustering algorithm to be soft?

Answer 10

Each object may belong to multiple clusters at once, with corresponding probabilities

Question 11

How do we define where data should go in the clustering process? Name one method.

Answer 11

A metric like similarity or distance, such as Euclidean, Manhattan, Jaccard distance or Jaccard similarity

Question 12

What is hierarchical clustering?

Answer 12

Clustering by partitioning data into a hierarchy at different levels

Question 13

What is a dendrogram?

Answer 13

A tree diagram that shows a hierarchy of clusters, where each node on the tree is a cluster

Question 14

What are the two methods of creating a dendrogram?

Answer 14

Agglomerative and divisive

Question 15

How can we create a dendrogram using an agglomerative method?

Answer 15

Starting with each item in its own cluster, we find the best pair to merge into a new cluster using a distance matrix, and repeat until all are fused together

Question 16

What is simple linkage between two clusters in an agglomerative dendrogram?

Answer 16

Where we define cluster distance as the distance between the two closest members in each cluster

Question 17

What is complete linkage between two clusters in an agglomerative dendrogram?

Answer 17

Where we define cluster distance as the distance between the two farthest members in each cluster

Question 18

What is average linkage between two clusters in an agglomerative dendrogram?

Answer 18

Where we define cluster distance as the average between all members in each cluster

Question 19

What is centroid linkage between two clusters in an agglomerative dendrogram?

Answer 19

Where we define cluster distance as the distance between the two centroids (worst, good, best) of each cluster

Question 20

What is Ward’s method between two clusters in an agglomerative dendrogram?

Answer 20

Where we join clusters only if it reduces the total distance from the centroids

Question 21

What is the downside to using simple linkage between two clusters in an agglomerative dendrogram?

Answer 21

Favours long, chain-like clusters

Question 22

What is the downside to using complete linkage between two clusters in an agglomerative dendrogram?

Answer 22

Breaks clusters into too many subclusters

Question 23

What is the downside to using average linkage between two clusters in an agglomerative dendrogram?

Answer 23

Makes a large amount of comparisons, so computationally slow

Question 24

What is the downside to using centroid linkage between two clusters in an agglomerative dendrogram?

Answer 24

Biased towards spherical clusters

Question 25

What is the downside to using Ward's method between two clusters in an agglomerative dendrogram?

Answer 25

Biased towards spherical clusters

Question 26

What are the two hyperparameters used in the DBSCAN algorithm?

Answer 26

Radius and minimum number of points

Question 27

When is a point a core point in DBSCAN?

Answer 27

If a point has more than the minimum number of points within its radius

Question 28

When is a point a border point in DBSCAN?

Answer 28

If a point has less than the minimum number of points within its radius, but is within the neighbourhood of a core point

Question 29

When is a point a noise point in DBSCAN?

Answer 29

If a point is not a core point or border point

Question 30

What are the benefits to using DBSCAN?

Answer 30

It is resistant to noise, and can handle weirdly shaped clusters

Question 31

What are the downsides to using DBSCAN?

Answer 31

It struggles with varying densities, and is hyperparameter reliant

Question 32

What are partitional clustering methods?

Answer 32

Methods that split our space into partitions, corresponding to clusters

Question 33

What is K-means clustering?

Answer 33

A partitional clustering method that uses distance to compute centroid clusters

Question 34

How does K-means clustering define clusters?

Answer 34

We initially choose a number of random data points to be initial cluster centroids, then iteratively recompute the centroids using some distance metric, and check for convergence

Question 35

What is the downside to K-means clustering?

Answer 35

It assumes that each cluster will be of similar size, struggling when they are not, plus the results depend on the starting centroids, meaning hyperparameter tuning must be performed

Question 36

What is the Gaussian Mixture Model (GMM)?

Answer 36

A partitional clustering method where each cluster is represented by a probability distribution

Question 37

What is clustering validation?

Answer 37

Using some metric to determine how successful clustering was

Question 38

What is an external validation method of clustering?

Answer 38

Measuring how clustering labels compare to externally supplied labels

Question 39

What is an internal validation method of clustering?

Answer 39

Measuring whether points that should be close actually are or not

Question 40

What is an relative validation method of clustering?

Answer 40

Comparing one clustering to another, and seeing if they agree

Question 41

What is cluster cohesion for clustering validation? Is it [internal/external/relative]?

Answer 41

A validation technique that measures the distance within a cluster. It is internal

Question 42

What is cluster separation for clustering validation? Is it [internal/external/relative]?

Answer 42

A validation technique that measures the distance between clusters. It is internal

Question 43

What is the silhouette coefficient for clustering validation? Is it [internal/external/relative]?

Answer 43

A validation technique that measures the mean distance to the nearest cluster minus the distance within clusters, being a mix between cohesion and separation. It is internal