linprog

Idea is to provide wrapper for most performant opensource optimization tools that are available. This shard is about linear and mixed integer optimization. According to http://plato.asu.edu/bench.html CLP is pretty good for linear programming, and other projects of COIN-OR (CBC and IPOPT) are fine for nonlinear.

CoinMP is decribed as simple API for CLP and CBC (just what I need!), but appears to be buggy and obsolete.

SYMPHONY looks easy enough to use and to distribute.

1. SYMPHONY works, but require few patches to be easily wrappable (namely extern C to disable mangling and #define printf to disable console spam). I've created fork https://github.com/konovod/SYMPHONY with patches
2. It is distributed under EPL1.0 that is incompatible with GPL, but otherwise pretty permissive.

Installation

1. Install COIN-OR SYMPHONY from package manager to get all dependencies
2. git clone https://github.com/konovod/SYMPHONY
3. configure && make
4. copy SYMPHONY/SYMPHONY/src/.libs/libSym.so.0.0.0 to /usr/lib, create symlink /usr/lib/libSym.so.1 pointing to it

Alternatively: you can use installation instructions for the upstream version: https://github.com/coin-or/SYMPHONY Master already has extern C fix but will spam some messages to console.

Usage

require "linalg"
require "linprog"

# linear programming
# x - solution, f - objective value
x, f = LinProg.solve(
  # c - objective function
  c: LA::GMat.new([[0, -1]]),
  # a_ub - left part of inequalities
  a_ub: LA::GMat.new([[-1, 1], [3, 2], [2, 3]]),
  # b_ub - right part of inequalities
  b_ub: LA::GMat.new([[1, 12, 12]]).t,
  # a_eq, b_eq - left and right parts of equalities, can be skipped if empty
  # bounds - constraints to variables
  bounds: {LinProg::Bound.none, LinProg::Bound.new(-3.0, Float64::INFINITY)})
pp x, f

# bound can be same for all vars
x, f = LinProg.solve(
  c: LA::GMat.new([[0, -1]]),
  a_ub: LA::GMat.new([[-1, 1], [3, 2], [2, 3]]),
  b_ub: LA::GMat.new([[1, 12, 12]]).t,
  bounds: LinProg::Bound.positive)
pp x, f

# tasks can use equalities, inequalities or both
x, f = LinProg.solve(
  c: LA::Mat.row([1, 7, 4, 2, 3, 5]),
  a_eq: LA::GMat.new([
    [1, 1, 1, 0, 0, 0],
    [0, 0, 0, 1, 1, 1],
    [1, 0, 0, 1, 0, 0],
    [0, 1, 0, 0, 1, 0],
    [0, 0, 1, 0, 0, 1],
  ]),
  b_eq: LA::Mat.column([30, 20, 15, 25, 10]))
pp x, f

# integer and mixed integer programming
x, f = LinProg.solve(
  c: LA::GMat.new([[0, -1]]),
  a_ub: LA::GMat.new([[-1, 1], [3, 2], [2, 3]]),
  b_ub: LA::GMat.new([[1, 12, 12]]).t,
  bounds: {LinProg::Bound.integer, LinProg::Bound.new(0.0, 6.0, integer: true)})
pp x, f

# there is also more complex interface that allows to save\load problems, but it's WIP, check spec dir for it

DSL

There is a dsl for nice creation of problems. It is WIP in a sense that more features will be added (most important is to check how it works with MultiArray to be on par with AMPL features), but right now following works:

  # Create a problem
  task = LinProg::SymbolProblem.new
  # add a vars with (optional) bounds
  x = task.create_var(bound: LinProg::Bound.integer)
  y = task.create_var(bound: LinProg::Bound.new(0.0, 6.0, true))
  # add constraints in a natural way
  task.st(x + 1 >= y)
  task.st(3*x + 2*y <= 12)
  task.st(2*x + 3*y <= 12)
  # set objective and direction of optimization
  task.maximize(y + 1)
  # solve
  task.solve
  # now vars have values from a solution
  x.value.should eq 1
  y.value.should eq 2

Roadmap:

• warmstarting
• sensitivity analysis
• access to solver parameters
• way to reset solver properly or otherwise avoid memory leak
• sparse matrix support
• DSL for problems creation
  • Check integration with https://github.com/konovod/multiarray
  • More fleshy example
• GLPK bindings to provide easy-to-install solution
• CI