[ library(branch_and_bound) | Reference Manual | Alphabetic Index ]

bb_min(+Goal, ?Cost, +Options)

Find one or all minimal solutions using the branch-and-bound method

Goal: The (nondeterministic) search goal
Cost: A (usually numeric domain) variable representing the cost
Options: A bb_options structure or variable

Description

A solution of the goal Goal is found that minimizes the value of Cost. Cost should be a variable that is affected, and eventually instantiated, by Goal. Usually, Goal is the search procedure of a constraint problem and Cost is the variable representing the cost. The solution is found using the branch and bound method: as soon as a solution is found, it gets remembered and the search is continued or restarted with an additional constraint on the Cost variable which requires the next solution to be better than the previous one. Iterating this process finally yields an optimal solution.

The possible options are

strategy:

continue (default): after finding a solution, continue search with the newly found bound imposed on Cost
restart: after finding a solution, restart the whole search with the newly found bound imposed on Cost
step: a synonym for 'restart'
dichotomic: after finding a solution, split the remaining cost range and restart search to find a solution in the lower sub-range. If that fails, assume the upper sub-range as the remaining cost range and split again.

The new bound (or the split point, respectively), is computed from the current best solution, taking into account the parameters delta and factor.

from:

number - an initial lower bound for the cost (default -1.0Inf). Only useful if Cost is not a domain variable.

to:

number - an initial upper bound for the cost (default +1.0Inf). Only useful if Cost is not a domain variable.

delta:

number - minimal absolute improvement required for each step (applies to all strategies). The default value of 1.0 is appropriate for integral costs. Any solution that improves on the best solution by less than this value will be missed.

factor:

number - minimal improvement ratio (with respect to the lower cost bound) for strategies 'continue' and 'restart' (default 1.0), or split factor for strategy 'dichotomic' (default 0.5)

solutions:

one (default): Compute one (of possibly multiple) optimal solutions.
all: Nondeterministically compute all optimal solutions. This has a performance penalty, as the search is restarted one more time after the optimum has been determined.

Note the dependence on the delta-parameter: the costs of these solutions may deviate by less than delta from the true optimum.

timeout:

number - maximum seconds of cpu time to spend (default: no limit)

report_success:

GoalPrefix/N - this specifies a goal to be invoked whenever the branch-and-bound process finds a better solution. GoalPrefix is a callable term (atom or compound) and N is an integer between 0 and 3. The invoked goal is constructed by adding N optional arguments to GoalPrefix: Cost, Handle and Module. Cost is a float number representing the cost of the solution found, Handle is a handle as accepted by bb_cost/2 or bb_solution/2, and Module is the context module of the minimisation. To disable any reporting, choose report_success:true/0. The default handler prints a message to log_output.

report_failure:

GoalPrefix/N - this specifies a goal to be invoked whenever the branch-and-bound process cannot find a solution in a cost range. GoalPrefix is a callable term (atom or compound) and N is an integer between 0 and 3. The invoked goal is constructed by adding N optional arguments to GoalPrefix: Cost, Handle and Module. Cost is a From..To structure representing the range of cost in which no solution could be found, Handle is a handle as accepted by bb_cost/2 or bb_solution/2, and Module is the context module of the minimisation. To disable any reporting, choose report_failure:true/0. The default handler prints a message to log_output.

report_timeout:

GoalPrefix/N - this specifies a goal to be invoked when the branch-and-bound process times out. GoalPrefix is a callable term (atom or compound) and N is an integer between 0 and 3. The invoked goal is constructed by adding N optional arguments to GoalPrefix: Cost, Handle and Module. Cost is a float number representing the cost of the best solution found, Handle is a handle as accepted by bb_cost/2 or bb_solution/2, and Module is the context module of the minimisation. To disable any reporting, choose report_timeout:true/0. The default handler prints a message to log_output.

The default options can be selected by passing a free variable as the Options-argument. To specify other options, pass a bb_options- structure in struct-syntax, e.g.

	    bb_min(..., ..., bb_options{strategy:dichotomic, timeout:60})

In order to maximize instead of minimizing, introduce a negated cost variable in your model and minimize that instead, e.g.

	    % maximize Profit
	    Cost #= -Profit,
	    bb_min(search(...), Cost, bb_options{}),

Modes and Determinism

bb_min(+, ?, +) is nondet

Modules

This predicate is sensitive to its module context (tool predicate, see @/2).

Fail Conditions

Goal has no solutions

Examples