linear programming · simplex method · L1-violation classifier
17 points are drawn from two overlapping clouds:
label +1 tends toward the upper left and label −1 toward the lower right. The clouds need not
be linearly separable. The program minimizes the L₁ norm of the nonnegative violation vector
η: minimize Σηᵢ, subject to
bᵢ(w₀ + wₓxᵢ + wᵧyᵢ) ≥ 1 − ηᵢ and ηᵢ ≥ 0.
The coefficients w₀,wₓ,wᵧ are unrestricted. The displayed boundary follows the
basic feasible solution through the primal-simplex pivots.
current LP solution
step trace
simplex pivot log