pybnb.pyomo.range_reduction

A Problem interface for implementing parallel range reduction on a PyomoProblem during a branch-and-bound solve.

Copyright by Gabriel A. Hackebeil (gabe.hackebeil@gmail.com).

class pybnb.pyomo.range_reduction.RangeReductionProblem(problem, best_objective=None, comm=None)

A specialized implementation of the pybnb.Problem interface that can be used to perform optimality-based range reduction on a fully implemented PyomoProblem by defining additional abstract methods.

listen(root=0)

Listen for requests to run range reduction. All processes within the communicator, except for the root process, should call this method.

Parameters:root (int) – The rank of the process acting as the root. The root process should not call this function.

sense()

Returns the objective sense for this problem.

Note

This method is abstract and must be defined by the user.

objective()

Returns a feasible value for the objective of the current problem state or self.infeasible_objective() if the current state is not feasible.

Note

This method is abstract and must be defined by the user.

bound()

Returns a value that is a bound on the objective of the current problem state or self.unbounded_objective() if no non-trivial bound is available.

Note

This method is abstract and must be defined by the user.

save_state(node)

Saves the current problem state into the given pybnb.node.Node object.

This method is guaranteed to be called once at the start of the solve by all processes involved to collect the root node problem state, but it may be called additional times. When it is called for the root node, the node.tree_depth will be zero.

Note

This method is abstract and must be defined by the user.

load_state(node)

Loads the problem state that is stored on the given pybnb.node.Node object.

Note

This method is abstract and must be defined by the user.

branch()

Returns a list of Node objects that partition the node state into zero or more children. This method can also be defined as a generator.

Note

This method is abstract and must be defined by the user.

notify_new_best_node(node, current)

Called when a branch-and-bound solver receives a new best node from the dispatcher. The Problem base class provides a default implementation for this method that does nothing.

Parameters:

• node (Node) – The new best node.
• current (bool) – Indicates whether or not the node argument is the currently loaded node (from the most recent load_state call).

notify_solve_finished(comm, worker_comm, results)

Called when a branch-and-bound solver finishes. The Problem base class provides a default implementation for this method that does nothing.

Parameters:

• comm (mpi4py.MPI.Comm or None) – The full MPI communicator that includes all processes. Will be None if MPI has been disabled.
• worker_comm (mpi4py.MPI.Comm or None) – The MPI communicator that includes only worker processes. Will be None if MPI has been disabled.
• results (SolverResults) – The fully populated results container that will be returned from the solver.

range_reduction_model_setup()

Called prior to starting range reduction solves to set up the Pyomo model

range_reduction_objective_changed(objective)

Called to notify that the range reduction routine has changed the objective

range_reduction_constraint_added(constraint)

Called to notify that the range reduction routine has added a constraint

range_reduction_constraint_removed(constraint)

Called to notify that the range reduction routine has removed a constraint

range_reduction_get_objects()

Called to collect the set of objects over which to perform range reduction solves

range_reduction_solve_for_object_bound(x)

Called to perform a range reduction solve for a Pyomo model object

range_reduction_model_cleanup()

Called after range reduction has finished to allow the user to execute any cleanup to the Pyomo model.

range_reduction_process_bounds(objects, lower_bounds, upper_bounds)

Called to process the bounds obtained by the range reduction solves