Quote:
Originally Posted by htlin
Hinted in my reply is that for PLA in particular, using any positive alpha gives you the same (equivalent) answer with exactly the same number of iterations. So convergencewise, alpha doesn't affect PLA at all. Not necessarily true for other algorithms, of course.

For the PLA, the line (defined by w0, w1, w2) only depends on the ratio, not the absolute value of the w's.
One thing, though. In my first PLA program I tested the sign by multiplying the two, and testing for >= 0. With all the w's 0, the value of W (dot) X is zero, which was >= 0. Changing the test to >0 fixed it.