Homework 2, Q7

[argentina]

I have run the perceptron with the linear regression weights as seeds and got 5 iterations on average for 10 points. This compares similarly with the 6 iterations I get with 0 as seed for each weight. On the other hand, when I run both of them with 100 points, I get surprising results: the linear regression + perceptron takes 43 iterations on average while the perceptron alone (with 0-valued weights as seeds) takes 35.

This sounds really counterintuitive to me, is this to be expected? Am I missing something?

Thanks!