Machine Learning - Clustering

CS102
Spring 2020

Clustering CS102
 Data Tools and Techniques
§ Basic Data Manipulation and Analysis
Performing well-defined computations or asking
well-defined questions (“queries”)
§ Data Mining
Looking for patterns in data
§ Machine Learning
Using data to build models and make predictions
§ Data Visualization
Graphical depiction of data
§ Data Collection and Preparation

Clustering CS102
 Machine Learning
Using data to build models and make predictions
Supervised machine learning
• Set of labeled examples to learn from: training data
• Develop model from training data
• Use model to make predictions about new data
Unsupervised machine learning
• Unlabeled data, look for patterns or structure
(similar to data mining)

Clustering CS102
 Clustering
Like classification, data items consist of values
for a set of features (numeric or categorical)
§ Medical patients
Feature values: age, gender, symptom1-severity,
symptom2-severity, test-result1, test-result2

§ Web pages
Feature values: URL domain, length, #images, heading1,
heading2, …, headingn

§ Products
Feature values: category, name, size, weight, price

Clustering CS102
 Clustering
Like classification, data items consist of values
for a set of features (numeric or categorical)
§ Medical patients Unlike classification,
Feature values: age, gender,
there is no label
symptom1-severity,
symptom2-severity, test-result1, test-result2

§ Web pages
Feature values: URL domain, length, #images, heading1,
heading2, …, headingn

§ Products
Feature values: category, name, size, weight, price

Clustering CS102
 Clustering
Like K-nearest neighbors, for any pair of data items
i1 and i2, from their feature values can compute
distance function: distance(i1,i2)
Example:
Features - gender, profession, age, income, postal-code
person1 = (male, teacher, 47, $25K, 94305)
person2 = (female, teacher, 43, $28K, 94309)
distance(person1, person2)

distance() can be defined as inverse of similarity()

Clustering CS102
 Clustering
GOAL: Given a set of data items, partition
them into groups (= clusters) so that items
within groups are close to each other based
on distance function
Ø Sometimes number of clusters is pre-specified
Ø Typically clusters need not be same size

Clustering CS102
 Some Uses for Clustering
§ Classification!
• Assign labels to clusters
• Now have labeled training data
for future classification
§ Identify similar items
• For substitutes or recommendations
• For de-duplication
§ Anomaly (outlier) detection
• Items that are far from any cluster

Clustering CS102
 K-Means Clustering
Reminder: for any pair of data items i1 and i2
have distance(i1,i2)
For a group of items, the mean value (centroid)
of the group is the item i (in the group or not)
that minimizes the sum of distance(i,i’) for all i’
in the group

Clustering CS102
 K-Means Clustering
For a group of items, the mean value (centroid)
of the group is the item i (in the group or not)
that minimizes the sum of distance(i,i’) for all i’
in the group
§ Error for each item: distance d from the mean
for its group; squared error is d 2
§ Error for the entire clustering:
sum of squared errors (SSE)

Remind you of anything?

Clustering CS102
 K-Means Clustering
Given set of data items and desired
number of clusters k, K-means groups the
items into k clusters minimizing the SSE

§ Extremely difficult to compute efficiently

Clustering CS102
 Clustering European Cities
By geographic distance, then by temperature

Clustering CS102
 Clustering European Cities
Distance = actual distance, k = 5

Clustering CS102
 Clustering European Cities
Distance = actual distance, k = 8, with cluster means

Clustering CS102
 Clustering European Cities
Distance = actual distance, k = 2, with cluster means

Clustering CS102
 Clustering European Cities
Distance = actual distance, k = 30

Clustering CS102
 Clustering European Cities
Distance = temperature, k = 5

Clustering CS102
 Clustering European Cities
Distance = temperature, k = 8, with means

Clustering CS102
 Clustering European Cities
Distance = temperature, k = 2

Clustering CS102
 Clustering European Cities
Distance = temperature, k = 3

Clustering CS102
 Clustering European Cities
Distance = temperature, k = 30

Clustering CS102
 Some Uses for Clustering
§ Classification
• Assign labels to clusters
• Now have labeled training data
for future classification
§ Identify similar items
• For substitutes or recommendations
• For de-duplication
§ Anomaly (outlier) detection
• Items that are far from any cluster

Clustering CS102