An open API service indexing awesome lists of open source software.

https://github.com/xuzijian629/pace2020

Exact Treedepth Solver based on Minimal Separator Enumeration
https://github.com/xuzijian629/pace2020

exact-algorithm treedepth

Last synced: over 1 year ago
JSON representation

Exact Treedepth Solver based on Minimal Separator Enumeration

Awesome Lists containing this project

README

          

# PACE 2020
This repository contains solvers for treedepth, submitted from team xuzijian629 at [PACE 2020](https://pacechallenge.org/2020/td/).

Authors: Zijian Xu, Dejun Mao and Vorapong Suppakitpaisarn from The University of Tokyo

UPD: We were ranked in [5th place](https://pacechallenge.org/2020/results/) in the competition.

UPD2: The official solver description is published at https://drops.dagstuhl.de/opus/volltexte/2020/13334/pdf/LIPIcs-IPEC-2020-31.pdf

UPD3: The conjecture used in the algorithm is resolved and the submitted solver is proved to be inexact. See https://arxiv.org/abs/2008.09822 for details.

UPD4: The slides presented at IPEC is [available](https://github.com/xuzijian629/pace2020/blob/master/IPEC.pdf).

# Algorithm
We enumerate minimal separtors to construct treedepth decomposition in a top-down way. A quite effective pruning is done by computing exact treedepth of many small components throughout the input graph.
## Solver description
A detailed explanation of our algorithm is provided [here](https://github.com/xuzijian629/pace2020/blob/master/submissions/solver-description-xuzijian629.pdf).

# Installation
- GCC version 5.4.0 or later

Our solver compiles source code, after checking the size of the input graph. So, just run `src/solver < input.gr`.

# Input format
Input graph format is specified [here](https://pacechallenge.org/2020/td/).

# Directories
- Source code for the solver is stored in `src/`.
- Other directories such as `benchmark/` and `test/` are for authors' personal use (local experiments).

# LICENSE
[MIT LICENSE](https://github.com/xuzijian629/pace2020/blob/master/LICENSE.md)