Clarification of Problem 3.1

[mileschen]

Actually, I could not understand what the learning task of the double semi-circle? It just said that there are two semi-circles of width thk with inner radius rad, separated by sep as shown (red -1 and blue is +1). Then, when we generate the 2000 examples, should we also give each example a corresponding color?

Could anyone possibly demonstrate this question clearly for me? Thanks very much!

[magdon]

The problem describes a geometrical region on the plane, half of which is shaded red and the other half blue. You generate 2000 points in this region randomly. If the point lands in the blue region, it is classified blue(+1) and otherwise red(-1). In this way you generate a training data set of 2000 points.

Does that clarify your confusion?

Originally Posted by mileschen

Actually, I could not understand what the learning task of the double semi-circle? It just said that there are two semi-circles of width thk with inner radius rad, separated by sep as shown (red -1 and blue is +1). Then, when we generate the 2000 examples, should we also give each example a corresponding color?

Could anyone possibly demonstrate this question clearly for me? Thanks very much!

[mileschen]

Yes, thank you very much. It is quite clear now.

[i_need_some_help]

I would like some clarification as well.

Does part (b) involve running the pocket algorithm (it says, in parentheses, "for classification") or would just running the 1-iteration linear regression algorithm suffice?

[i_need_some_help]

Originally Posted by i_need_some_help

I would like some clarification as well.

Does part (b) involve running the pocket algorithm (it says, in parentheses, "for classification") or would just running the 1-iteration linear regression algorithm suffice?

I made a mistake and was running regression with input prepared for regression, not classification. No need for pocket.