 |
|
|
 |
|
|
|
|
twizansk |
| | | | | | posted 6/29/2010 00:00 |   | | | | | | | | | I have a question about the implementation of a support vector machine for the non-separable case.
In this case, the usual constraints on the lagrange multipliers become:
0 <= alpha_i <= C
where C is the coefficient of the slack variable term in the objective function.
We then solve the optimization problem, use the KKT conditions to find w and use the KKT conditions for any point for which alpha < C, to find the offset parameter b.
My question is, what do we do if all alphas equal C. How do we find b in that case?
Thanks
| | |
|
|
|
|
|
