Email from professor.
> We have two concerns regarding the second homework assignment.
>
> 1. Our S-K training algorithm converged and we obtained our alpha values. It is unclear to me how to compute our output function. Is the output function g(x) = sign( sigma(alpha_i *y_i * K(x, x'i) +(B-A)/2)?
>
Yes.
> 2. Also, we are unsure of how to apply lambda during the test case. How do we scale a test input using lambda if we don't know whether to apply it to m+ centroid or m- centroid? Since the kernel function is K(x, x'i), am I correct to assume that lambda is not applied to the test input vector x?
>
Like you assumed, you don’t scale the test input vector.
(
Edited: 2017-10-13)
Email from professor.
> We have two concerns regarding the second homework assignment.
>
> 1. Our S-K training algorithm converged and we obtained our alpha values. It is unclear to me how to compute our output function. Is the output function g(x) = sign( sigma(alpha_i *y_i * K(x, x'i) +(B-A)/2)?
>
Yes.
> 2. Also, we are unsure of how to apply lambda during the test case. How do we scale a test input using lambda if we don't know whether to apply it to m+ centroid or m- centroid? Since the kernel function is K(x, x'i), am I correct to assume that lambda is not applied to the test input vector x?
>
Like you assumed, you don’t scale the test input vector.