[Previous entry: "My *NIX Journal"] [Main Index] [Next entry: "Random links"]
10/04/2002 Entry: "Work related stuff..."
This is my work blog. I'd rather if you left this on alone :)
Replies: 4 comments
Progress report: gridplot
I've hand-verified the sum-across-row output. Next steps include:
1. sum across column
2. verification of sum across column
3. automation of feeding the data into NBD
Posted by Fahd @ 10/31/2002 12:35 PM EST
Hamming distance may have some applicability in grid vectors. Maybe we can differentiate based upon the Hamming distance of the grid plot vectors within a user from the user's mean stroke's gridplot vector. Okay, so that only looks at how far apart each stroke is from the user's average behavior. Nevertheless, if we can feed Hamming distance to anything that feature-selects on the basis of reduction of entropy, maybe a neural network, it may add something to the results.
Posted by Fahd @ 10/31/2002 12:33 PM EST
So when I ran the Naive Bayes Detector on an occupancy grid of users' character strokes, the results turned out to be pretty poor. One option is to increase the resolution of the grid and then sum across rows or columns, and run the NBD on the resulting vector. Another is to create a "mean" stroke, or a best fit stroke, and plot that instead. Maybe then the occuancy grid idea would work better.
Posted by Fahd @ 10/08/2002 10:52 AM EST
Roy asked me to think about how features can be automatically extracted from a given dataset. We discussed earlier that any such extractor would have to be given some "hint" as to what the nature of the data was, e.g. temporal, sequential, etc.
Posted by Fahd @ 10/08/2002 10:45 AM EST
Page Views since April 2001: