picos.reforms.reformulation

Backend for problem reformulation classes.

Classes

class picos.reforms.reformulation.Reformulation(theObject)

Bases: abc.ABC

Base class for problem reformulations.

Abstract base class for a reformulation from one (possibly already reformulated) problem form to another.

__init__(theObject)

Initialize Reformulation instances.

Parameters

theObject (Problem or Reformulation) – The input to work on; either an optimization problem or the (future) output of another reformulation.

abstract backward(solution)

Translate back a solution from reformulated to original problem.

Transforms a single Solution to output to a solution of input.

The method is allowed to modify the solution; it is not necessary to work on a copy. In particular, attach_to can be used if forward has created a deep copy of the problem.

execute()

Reformulate the problem and obtain a solution from the result.

For this to work there needs to be a solver instance at the end of the reformulation pipeline, which would implement its own version of this method that actually solves the problem and produces the first solution.

abstract forward()

Perform the initial problem reformulation.

Creates a modified copy or clone of the problem in input and stores it as output.

See copy and clone for the differences between a copy and a clone.

Implementations are supposed to do the necessary bookkeeping so that backward can transform a solution to the new problem back to a solution of the original problem.

abstract classmethod predict(footprint)

Predict the reformulation’s effect on a problem footprint.

Given a problem footprint, returns another problem footprint that a problem with the former one would be reformulated to.

This is used to predict the effects of a reformulation when planning a solution strategy without the cost of actually transforming a problem.

reset()

Reset the pipeline from this reformulation onward.

This is done whenever a reformulation does not implement update so that succeeding reformulations do not attempt to update a problem which was completely rewritten as this may be inefficient.

abstract classmethod supports(footprint)

Whether the reformulation affects problems with the given footprint.

The reformulation must support every problem with such a footprint and the resulting problem should have a changed footprint.

abstract update()

Update a previous problem reformulation.

Updates output and related bookkeeping information with respect to changes in input.

Raises

NotImplementedError – If performing an update is not feasible for the reformulation.

property input

The input problem.

output

The output problem.

successor

The next reformulation in the pipeline.

property verbosity

Verbosity level of the reformulation; same as for input problem.