User Guides#
The usage guide works one example through from start to finish: calibrating survey weights, from the mathematics and a scipy baseline to a fast Cvxium solve. Along the way it writes a new objective and a new constraint layer, explains how Cvxium assembles Hessian structure, and covers barrier initialization, caching, logging, Pareto frontiers, and conic and saddle-point problems.
The guide to large linear programs solves a media plan with 20,000
variables and 50 constraints, compares Cvxium with scipy’s linprog,
and explains the structure and the settings behind the speedup.
The guide to banded quadratic programs smooths 100,000 nonnegative power readings with a tridiagonal quadratic, compares Cvxium with a steel-manned cvxpy across five solvers, and shows how a band structure carries through every Newton step.