Skip to content

Heuristic algorithms for solving the budget-constrained prize collecting Steiner forest problem.

License

Notifications You must be signed in to change notification settings

amritagupta/budget-pcsf-semigradient-ascent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

17 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

budget-pcsf-semigradient-ascent

Heuristic algorithms for solving the budget-constrained prize collecting Steiner forest problem.

  • Citation Info: Gupta, A. and Dilkina, B., 2019, November. Budget-Constrained Demand-Weighted Network Design for Resilient Infrastructure. In 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI). IEEE.

Overview

We address the problem of allocating a fixed budget towards restoring edges to maximize the satisfied travel demand between locations in a network, which we formalize as the budget-constrained prize-collecting Steiner forest problem. We prove that the satisfiable travel demand objective exhibits restricted supermodularity over forests, and utilize this property to design an iterative algorithm based on maximizing successive modular lower bounds for the objective that finds better solutions than a baseline greedy approach. We also propose an extremely fast heuristic for maximizing modular functions subject to knapsack and graph matroid constraints that can be used as a subroutine in the iterative algorithm, or as a standalone method that matches the greedy baseline in terms of quality but is hundreds of times faster. We evaluate the algorithms on synthetic data, and apply them to a real-world instance of retrofitting the Senegal national and regional road network against flooding.


Setup

Dependencies

All code was written in Python 2.7.6.

Directory Structure

|____data
| |____synthetic_graphs
| | |____planar
| | | |____demands
| | | | |____G_n_edges_500_instance_0_demand_data.p
| | | | |____...
| | | |____edge_costs
| | | | |____G_n_edges_500_instance_0_cost_data.p
| | | | |____...
| | | |____edge_max_values_20
| | | | |____G_n_edges_500_instance_0_edge_values_data.p
| | | | |____...
| | | |____edge_values
| | | | |____G_n_edges_500_instance_0_edge_values_data.p
| | | | |____...
| | | |____graphs
| | | | |____G_n_edges_500_instance_0.graphml
| | | | |____...
|____LICENSE
|____notebooks
| |____generate-random-erdos-renyi-graph.ipynb
| |____generate-random-planar-graph.ipynb
|____output
| |____synthetic_graphs
| | |____er
| | |____planar
|____README.md
|____scripts
| |____budgetpcsf.py
| |____computemaxval.py
| |____run_experiments.py

How To Run

  1. The .ipynb notebooks in notebooks/ generate synthetic instances based on either Erdos-Renyi graphs or planar graphs. They populate the data/ folder with graphs, random edge costs and demand between pairs of nodes, as well as the demand between endpoints of each edge (edge values), which are used by the optimization algorithms.

  2. scripts/computemaxval.py estimates the maximum value of each edge by sampling 20 random spanning trees and evaluating the reduction in satisfiable travel demand when the edge is removed. Results are written to data/.

  3. scripts/budgetpcsf.py implements:

    a. greedy

    b. knapsack-repair

    c. semigradient ascent (using either greedy or knapsack-repair as subroutines)

    It also contains methods called by scripts/computemaxval.py and helper functions.

  4. scripts/run_experiments.py shows how we set up all experiments in our ICTAI 2019 paper.


Results

Random Planar Graphs

Relative Solution Quality

Runtime Performance

Random Erdos-Renyi Graphs

Relative Solution Quality

Runtime Performance