[kundanjan]

Summary

Implements Support Vector Machine classification (SVC) and regression (SVR) with a pure-Java SMO (Sequential Minimal Optimization) solver.

Components

• Kernel — linear, polynomial, RBF, sigmoid with sklearn-compatible gamma scaling/auto
• SMOSolver — Platt's SMO for the C-SVC dual QP with second-order working set selection
• SVC — binary classification + one-vs-one multi-class (6 tests)
• SVR — epsilon-insensitive regression via SMO-style solver (5 tests)

Key Details

• Decision function uses sum(αᵢyᵢK(xᵢ,x)) + b convention
• Multi-class uses sklearn's one-vs-one strategy with voting
• SVR uses β = α - α* formulation with ε-tube
• All tests pass with checkstyle clean