Problem 3.4 (and 3.3)
Jacob
Im wondering, are we supposed to use svmtrain or my_svm on this
problem?
Is the degree of the polynomial kernel up to us to choose?
Another question, I understand if you dont want to give this one away:
If running QP on problem 3.3 results in the maximum number of
iterations being exeeded, or the program keeps running for a
reasonalbe amount of time without producing a result, can I then
assume I have a fault in my QP, or is it possible that matlab's
implementation of QP (or svmtrain with qp option) does not cope with
the problem?
/Jacob
Vinay Jethava
On Feb 14, 2012, at 5:26 PM, Jacob wrote:
> Hi!
>
> Im wondering, are we supposed to use svmtrain or my_svm on this
> problem?
> Is the degree of the polynomial kernel up to us to choose?
In 3.4, You can use either my_svm or svmtrain (e.g. see how much is the difference between the two cases.)
In either case, you need to use the kernel in part 2 (b) i.e. K(z, x) = (1+ x^\top z)^2
On a side note, what happens if you use a better polynomial function e.g. svmtrain with default poly kernel?
> Another question, I understand if you dont want to give this one away:
> If running QP on problem 3.3 results in the maximum number of
> iterations being exeeded, or the program keeps running for a
> reasonalbe amount of time without producing a result, can I then
> assume I have a fault in my QP, or is it possible that matlab's
> implementation of QP (or svmtrain with qp option) does not cope with
> the problem?
HINT: As you said I can’t give this one away :) However, consider the following: why is SVM and SMO a big thing - since quadprog does solve the same problem.
> /Jacob