Clarification Required for PLA problem approach

[pranav_s]

According to the statement (as far as i could infer) I :
1.Create a database of lets say 1000 points and apply the target function (randomly chosen) to get their respective o/p(+1/-1)

2. in question 7, N=10, I pick up 10 random points (1 sample set) from the created database and get my 'g' using the PLA. Now, since I have a 'g' on which points do I apply it and then calculate Pr[f!=g] exactly??
As mentioned I perform the same thing on 1000 sample sets. and calculate Average Pr and Average number of iterations to converge. So, essentially I am converging to 1000 different 'g's. Is this right?

similarly for N=100??
Please, correct if there is any discrepancy.

[yaser]

Originally Posted by pranav_s

Now, since I have a 'g' on which points do I apply it and then calculate Pr[f!=g] exactly?

On out-of-sample points, i.e., fresh points generated according to the same (uniform) distribution. You can generate as many of these points as you want to get better accuracy. If you want an exact answer, calculate the area where and divide by the total area (which is 4).

[miguel.penalver@mail.com]

Originally Posted by pranav_s

As mentioned I perform the same thing on 1000 sample sets. and calculate Average Pr and Average number of iterations to converge. So, essentially I am converging to 1000 different 'g's. Is this right?

similarly for N=100??

Right.

[diethealthcare]