Revision as of 12:00, 12 April 2012

Third Homework, ECE662 Spring 2012

Email code to your instructor before 11:59pm, Friday April 27, 2012. Report due before 11:59pm, Monday April 30, in your instructor's Rhea dropbox. Make sure to drop it in the correct homework folder!!!!. It is the one at the very bottom of the page.

Automatic Pattern Recognition Contest!

An anonymous company has agreed to share real data with us, so we are going to have a little contest using this data! The data comes from a five-class classification problem using 13 features. We are looking for the student who will design the most accurate classifier using this data.

The training data consists of 550 data points (i.e. 550 points in a 13 dimensional space) along with the correct label for each point. Use this data, along with any method of your choice, to design what you think is an accurate classifier. When you are done designing your classifier, email your source code to your instructor, and you will receive the testing data. Then without changing your code, test your classifier on the testing data and note its accuracy. Summarize your method and results in a report.

If you feel like sharing your results and methods publicly, feel free to post a copy of your work below, but only after the deadline for the homework has past.

The discussion page for this homework is here.

Grading NEW!

If you thought this course could not been more fun, then you are wrong. With this competition, everything becomes more interesting. We are talking about a competition among the best students in the world in one of the coolest field of study, pattern recognition! Which classifier will make a better prediction for this data, SVM, Bayes, KNN .. ?? We will know soon.

Everything seems to be perfect except the judging. This is because we are our own judges. It is taugh to be rough on yourself. Therefore, I propose to do something, we will send professor our classifier (code) when we finish, then, professor will give us the test data without the answer to the predictions, so instead of providing our score, we will provide our answers (see below). Then on April 30th, a script will grade each one of us, and will find the winner. I volunteer to write the script that will calculate in one run the success of each individual classifier.

The final file we respond should look like: (the first line is just the name of your classifier).

KNN with K=24
1
2
1
2
3
4
2
3
1
2
1
2
1

Professor will confirm in class if we find the "scores" in this way.
Questions? Comments?
Daniel Rugeles

Student Reports

• link to a report
• link to another report.