SVM primal formulation, does the constants constraints matter/

111 Views Asked by At

When finding the maximum margin separator in the primal form we have the quadratic program:

$$min\frac{1}{2}||\theta||^2$$ $$\text{ subject to: } y^{(t)}(\theta \cdot x^{(t)} + \theta_0) \geq 1, \ t=1,...,n$$

Saying basically to find the maximum margin separator. The margin size will be:

$$\frac{1}{||\theta||}$$

does the size of the margin change whether we change the constant of the constraints?

i.e. if we have:

$$\text{ subject to: } y^{(t)}(\theta \cdot x^{(t)} + \theta_0) \geq k, \ t=1,...,n$$

instead of 1?

If it does not matter, why doesn't this matter? How is it an equivalent formulation regardless of the exact constant for the constraint?

1

There are 1 best solutions below

0
On

The constraints are homogeneous in the triple $(θ,θ_0,1)$ resp. $(θ,θ_0,k)$, so yes, multiplying this triple with a constant will still have the same separating plane as the solution, however, the margin size will have to be computed as $\frac{k}{\|θ\|}$.

That is why there exists a second formulation where the separation number $k$ is maximized while the hyperplane normal $θ$ stays normed, $\|θ\|=1$.