Homework 3 [ETC 3 hours]
Due Thursday, September 5, 2007

Reactions [15 minutes]

1. Read the blogs of other students in the class. Comment on at least 3 of them.

[Anywhere] Reading (ETC: 1 hour):

1. Read: Allen Chris Long, `Those look similar!' issues in automating gesture design advice (ETC: 10 minutes)
2. Read: Allen Chris Long: Visual similarities (ETC: 30 minutes)
3. Blog 4: Write Summary and Discussion for Reading (ETC: 20 minutes) - Note, you are free to write either one for both or two/one for each if you want to keep them separate on your blog. They are on the same topic. The first one is a short 4 page overview of the second longer 8 page paper that goes into the details.

Coding [ETC 2 hours]:

1. Code up the linear classifier as specified by Rubine. (Make sure to code it in a object orientated way...)[ETC 1.5 hr]
Note: You will want to have a matrix package:
• JAMA for Java: http://math.nist.gov/javanumerics/jama/
• or untested for C++: http://www.techsoftpl.com/matrix/download.htm
• or untested for C++: http://www.programmersheaven.com/download/30784/download.aspx
2. Set up your linear classifier to recognize using the Rubine features.[ETC 3 minutes - since already coded in previous hw]
3. Compute the accuracy of your Rubine feature-based classifier on the math data.[ETC 5 minutes - just printing data]
4. Bring in the classification for each symbol, to be ready to discuss which examples did not work and why [ETC 10 minutes]
5. Code up the extra features suggested by Long. [ETC: 20 minutes]
6. Compute the accuracy of your Long feature-based classifier on the math data [ETC 2 minutes]
7. Bring in the classification for each symbol, to be ready to discuss which examples did not work and why [ETC 10 minutes]
8. Note that for this homework, to compute your results you are simply rerunning each sample of your training data on your linear classifier. (Additional test data will be provided to you later.)