-
-
Courses
Find courses by:
Collections
Cross-Disciplinary Topic Lists
Translated Courses
-
About
-
Donate
-
Featured Sites
 |
This lecture covers hierarchical clustering and introduces k-means clustering. This image is from the Wikimedia Commons. This image is in the public domain. |
About this Video
Topics covered: Feature vectors, scaling, k-means clustering.
Resources
- Lecture code handout (PDF)
- Lecture code (PY)
- Lecture slides (PDF)
- Lecture data files (ZIP) (This ZIP file contains: 3 .txt files.)
Recitation 8: Hierarchical and k-means Clustering
About this Video
Topics covered: Unsupervised learning, k-means clustering, distance metric, cluster merging, centroid, k-mean error, holdout set, k value significance, features of k-means clustering, merits and disadvantages of types of clustering.
How do we use nominal (non-numeric or noncontinuous) categories as features?
› View/hide answer
Convert each possible value to a real number.
Why do we need to use scaling (normalization)?
› View/hide answer
To indicate the relative importance of each feature.
How does k-means clustering work?
› View/hide answer
A number 'k' points are chosen, randomly or otherwise, to be the initial centroids; all other points are assigned to their nearest centroid. A new, better centroid is then chosen for each cluster, and we rinse and repeat until the difference between our current set of clusters and the previous set is insignificant.
At an institute of higher education that shall remain nameless, it used to be the case that a human adviser would help each student formulate a list of subjects that would meet the student's objectives. However, because of financial troubles, the Institute has decided to replace human advisers with software. Given the amount of work a student wants to do, the program returns a list of subjects that maximizes the amount of value. The goal of this problem set is to implement optimization algorithms.
Problem set 10 is assigned in this session. The instructions and solutions can be found on the session page where it is due, Lecture 22 Using Graphs to Model Problems, Part 2.
MIT OpenCourseWare makes the materials used in the teaching of almost all of MIT's subjects available on the Web, free of charge. With more than 2,000 courses available, OCW is delivering on the promise of open sharing of knowledge.
© 2001–2012
Massachusetts Institute of Technology
Your use of the MIT OpenCourseWare site and materials is subject to our Creative Commons License and other terms of use.
