This gem is using Cbc, an Integer Linear Programming Library, to provide optimization problems solving to ruby. With Ruby-Cbc, you can model you problem, solve it and find conflicts in case of infeasibility.
It uses the version 2.9.9 of Cbc, and requires the version 2.9.9 of gem cbc-wrapper.
You need to have cbc
installed for ruby-cbc
to work.
- On a mac, you can execute
brew install cbc
- On Debian and Ubuntu, use
apt-get install coinor-libcbc-dev
- On Archlinux, use
pacman -S coin-or-cbc
- On docker you can check the dockerfiles in test/installation-tests/
Add this line to your application's Gemfile:
gem 'ruby-cbc'
And then execute:
$ bundle
Or install it yourself as:
$ gem install ruby-cbc
On Heroku, you will need to tweak your installation a bit: you can install the cbc library with the Apt buildpack with an Aptfile of:
coinor-libcbc-dev
You will need to set LD_LIBRARY_PATH so it can find LAPACK and BLAS (see this issue).
heroku config:set LD_LIBRARY_PATH=/app/.apt/usr/lib/x86_64-linux-gnu/lapack:/app/.apt/usr/lib/x86_64-linux-gnu/blas
# The same Brief Example as found in section 1.3 of
# glpk-4.44/doc/glpk.pdf.
#
# maximize
# z = 10 * x1 + 6 * x2 + 4 * x3
#
# subject to
# p: x1 + x2 + x3 <= 100
# q: 10 * x1 + 4 * x2 + 5 * x3 <= 600
# r: 2 * x1 + 2 * x2 + 6 * x3 <= 300
#
# where all variables are non-negative
# x1 >= 0, x2 >= 0, x3 >= 0
#
m = Cbc::Model.new
x1, x2, x3 = m.int_var_array(3, 0..Cbc::INF)
m.maximize(10 * x1 + 6 * x2 + 4 * x3)
m.enforce(x1 + x2 + x3 <= 100)
m.enforce(10 * x1 + 4 * x2 + 5 * x3 <= 600)
m.enforce(2 * x1 + 2 * x2 + 6 * x3 <= 300)
p = m.to_problem
p.solve
unless p.proven_infeasible?
puts "x1 = #{p.value_of(x1)}"
puts "x2 = #{p.value_of(x2)}"
puts "x3 = #{p.value_of(x3)}"
end
Let's have a model :
model = Cbc::Model.new
3 variable kinds are available:
x = model.bin_var(name: "x") # a binary variable (i.e. can take values 0 and 1)
y = model.int_var(L..U, name: "y") # a integer variable, L <= y <= U
z = model.cont_var(L..U, name: "z") # a continuous variable, L <= z <= U
Name is optional and used only when displaying the model.
If you don't specify the range, the variables are free.
You can also use Cbc::INF
as the infinity bound.
Each one of these 3 kinds have also an array method that generate several variables. For instance to generate 3 positive integer variables named x, y and z :
x, y, z = model.int_var_array(3, 0..Cbc::INF, names: ["x", "y", "z"])
You can enforce constraints:
model.enforce(x + y - z <= 10)
You are not restricted to usual linear programming rules when writing a constraint.
Usually you would have to write x - y = 0
to express x = y
.
Ruby-Cbc allows you to put variables and constants on both sides of the comparison operator. You can write
model.enforce(x - y == 0)
model.enforce(x == y)
model.enforce(x + 2 == y + 2)
model.enforce(0 == x - y)
Ruby-Cbc allows you to name your constraints. Beware that their name is not an unique id. It is only a helper for human readability, and several constraints can share the same function name.
model.enforce(my_function_name: x + y <= 50)
model.constraints.map(&:to_function_s) # => ["my_function_name(x, y)"]
Linear constraints are usually of the form
a1 * x1 + a2 * x2 + ... + an * xn <= C
a1 * x1 + a2 * x2 + ... + an * xn >= C
a1 * x1 + a2 * x2 + ... + an * xn == C
With Ruby-Cbc you can write
2 * (2 + 5 * x) + 4 * 5 + 1 == 1 + 4 * 5 * y
The (in)equation must still be a linear (in)equation, you cannot multiply two variables !
You can set the objective:
model.maximize(3 * x + 2 * y)
model.minimize(3 * x + 2 * y)
the Model
instances have a to_s
method. You can then run
puts model
The model will be printed in LP format.
For instance:
Maximize
+ 10 x1 + 6 x2 + 4 x3
Subject To
+ x1 + x2 + x3 <= 100
+ 10 x1 + 4 x2 + 5 x3 <= 600
+ 2 x1 + 2 x2 + 6 x3 <= 300
Bounds
0 <= x1 <= +inf
0 <= x2 <= +inf
0 <= x3 <= +inf
Generals
x1
x2
x3
End
To solve the model, you need to first transform it to a problem.
problem = model.to_problem
You can define a time limit to the resolution
problem.set_time_limit(nb_seconds)
You can solve the Linear Problem
problem.solve
You can specify arguments that match the cbc command line
problem.solve(sec: 60) # equivalent to $ cbc -sec 60
problem.solve(log: 1) # equivalent to $ cbc -log 1
For more examples of available options, if coinor-cbc
is installed run
$ cbc
then type ?
Once problem.solve
has finished you can query the status:
problem.proven_infeasible? # will tell you if the problem has no solution
problem.proven_optimal? # will tell you if the problem is solved optimally
problem.time_limit_reached? # Will tell you if the solve timed out
To have the different values, do
problem.objective_value # Will tell you the value of the best objective
problem.best_bound # Will tell you the best known bound
# if the bound equals the objective value, the problem is optimally solved
problem.value_of(var) # will tell you the computed value or a variable
Sometimes a problem has no feasible solution. In this case, one may wonder what is the minimum subset of conflicting inequations. For this prupose, you can use
problem.find_conflict # Will return an array of constraints that form an unsatifiable set
problem.find_conflict_vars # Will return all variables involved in the unsatisfiable minimum set of constraints
It finds a minimum subset of constraints that make the problem unsatisfiable. Note that there could be several of them, but the solver only computes the first one it finds. Note also that it does so by solving several instances of relaxed versions of the problem. It might take some time! It is based on QuickXplain (http://dl.acm.org/citation.cfm?id=1597177).
One way to see the results nicely could be
problem.find_conflict.map(&:to_function_s)
Bug reports and pull requests are welcome on GitHub at https://github.com/gverger/ruby-cbc.