Support Vector Machine
Inspect decision scores, margins, hinge loss, kernels, and active support influence.
SVM learns a margin-based decision function. ModelOrigen records real optimizer checkpoints and exposes decision scores, hinge loss, objective, margin evidence, kernel behavior, and active coefficients.
Step by step
- Encode the two classes with opposite signs.
- Choose a kernel and C penalty.
- Optimize coefficients and bias under the margin objective.
- Identify samples with non-zero support influence.
- Calculate the query decision score.
- Use the score sign for the final class and inspect the margin/boundary.
Core formulas
Decision function
f(x) = Σ αᵢ yᵢ K(xᵢ,x) + bSupport coefficients and the kernel define the score.
Hinge loss
max(0, 1 − y·f(x))Penalizes samples that violate the desired margin.
Soft-margin objective
½||w||² + C Σ hingeᵢC balances a wider margin against violations for the linear intuition.
When to use Support Vector Machine
- Margin separation is useful and data size is manageable.
- High-dimensional features may be effective.
- A kernelized nonlinear boundary is worth the extra complexity.