Contents
Instruction for Peer Review of HW1, ECE662, Spring 2012
If you go to your own dropbox and click on "my assignments", you should find an anonymous homework waiting to be reviewed. There should be a comment box below the homework, as well as a pdf upload button.
Part 1
Provide detailed comments on the problem addressed, experiments, conclusions, and report.
- Summarize what was done and how it was done.
- Comment on the "good" things in the report
- Comment on what could be improved, and how to improve it. (Phrase things nicely. Be diplomatic!)
You can write your comments directly in the comment box, or upload a pdf.
Part 2
Assign a grade out of 100 points and write this grade on the top line of the comment box. Your points should be divided as follows.
35 Points: Problem definition and statement
Is the problem/question investigated concerned with a relevant aspect of "classification assuming normally distributed features"? Is the problem/question addressed clearly stated? Is the problem/question investigated interesting and extensive enough. (If the writing is so poor that you have no idea what was done, feel free to take off a large number of points, or even all 35 points.
Take off 15 points if only the "Naive Bayes" approach was tested (i.e. if the features were assumed to be independent).
35 Points: Experiments
Are the experiments relevant to the problem investigated? Are there enough experiments (to investigate the problem and be able to conclude)? Are the axes of all graphs and plots clearly labeled? Do all graphs and plots have a title? (If the writing is so poor that you have no idea what was done, feel free to take off a large number of points, or even all 35 points.)
20 Points: Conclusions
Are the conclusions clearly stated? Are the conclusions supported by the experiments? Are the conclusions interesting? Note that a negative conclusion, such as "this does not work", can still be interesting. (If the writing is so poor that you have no idea what was done, feel free to take off a large number of points, or even all 20 points.)
10 points: Presentation
Questions/comments
Feel free to write your questions and comments below.
- How many points should we take off if the axes of the plots are not labeled?
- You will have to decide this yourself. Try to come up with a reasonable scheme, and specify what scheme you are using when giving your grade. -pm
- I have a question concerning "Naive Bayes": When someone used only a diagonal covariance matrix, did he or she automatically only do Naive Bayes? Or is it more about the question how the person calculates the probability density function? So, if the equation (or matlab function) is able to work with non-independent features than it would be fine and I don't have to take off the 15 points?
- The answer to your first question is: No. In the two category case, one can always diagonalize the covariance matrices. So I don't have a problem with people working with diagonal matrices. Naive Bayes is when one is using real data and just assuming that the covariance matrices of the classes are diagonal (without even checking that it is true): this immediately simplifies the problem to a sequence of 1D problems, thus pretty much trivializing the whole thing. -pm
- Write a question here.
- Answer here
