+ All Categories
Home > Documents > SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger...

SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger...

Date post: 18-Dec-2015
Category:
View: 215 times
Download: 0 times
Share this document with a friend
Popular Tags:
35
SATzilla-07: The Design SATzilla-07: The Design and Analysis of an and Analysis of an Algorithm Portfolio for Algorithm Portfolio for SAT SAT Lin Xu, Frank Hutter, Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton- Holger H. Hoos and Kevin Leyton- Brown Brown University of British Columbia University of British Columbia {xulin730, hutter, hoos, kevinlb}@cs.ubc.ca {xulin730, hutter, hoos, kevinlb}@cs.ubc.ca
Transcript
Page 1: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla-07: The Design and SATzilla-07: The Design and Analysis of an Algorithm Analysis of an Algorithm

Portfolio for SATPortfolio for SAT

Lin Xu, Frank Hutter,Lin Xu, Frank Hutter,Holger H. Hoos and Kevin Leyton-Brown Holger H. Hoos and Kevin Leyton-Brown

University of British ColumbiaUniversity of British Columbia

{xulin730, hutter, hoos, kevinlb}@cs.ubc.ca{xulin730, hutter, hoos, kevinlb}@cs.ubc.ca

Page 2: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

2

Outline

Motivation

History of SATzilla and related work

SATzilla methodology

Example Problem

SATzilla for the SAT Competition

Conclusions and ongoing research

Page 3: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

Motivation

Lots of high performance solvers, but …. No single SAT solver dominates all others on

all types of instances

Question: How to select the best solver for a given SAT instance?

Page 4: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

4

Algorithm Selection Problem [Rice, 1976]

Reference: Select solvers based on previous experience or research papers

“Winner-Take-All”: Test solvers on some samples from target distribution;

select the solver with best performance.

SATzilla: Automatically based on instance characteristics’

Page 5: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

Related work: Portfolio of stochastic algorithm [Gomes & Selman,1997]

Running multiple algorithms at the same time

Reinforcement learning [Lagoudakis & Littman, 2001]

Select branching rule at each decision point

Branch & bound algorithm selection [Lobjois & LemaÎter, 1998]

Based on an estimation of search tree size

Page 6: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

6

History of SATzilla Old SATzilla [Nudelman, Devkar, et. al, 2003 ]

2nd Random 2nd Handmade (SAT) 3rd Handmade

SATzilla-07 1st Handmade 1st Handmade (UNSAT) 1st Random 2nd Handmade (SAT) 3rd Random (UNSAT)

Page 7: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

7

SATzilla-07 Methodology (offline)

Target

Distribution

Solvers

Features

Final solver

selection

Build Model

Winner-

Take-All

Collect Data

Pre-Solvers

Page 8: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

8

SATzilla-07 Methodology (online)

Run

Pre-solver

Compute

Features

Predict

Runtime

Run Winner-

Take-All

Run Best

If error

If error AND time left; run second best

Page 9: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

9

Solvers Used Eureka [Nadel, Gordon, Palti & Hanna,

2006]

Kcnfs2006 [Dubois & Dequen, 2006]

March_dl2004 [Heule & Maaren, 2006]

Minisat2.0 [Eén & Sörensson, 2006]

OKsolver [Kullmann, 2002]

Rsat [Pipatsrisawat & Darwiche, 2006]

Vallst [Vallstrom, 2005]

Zchaff_Rand [Mahajan, Fu & Malik, 2005]

Page 10: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla-07 Example

Using quasi-group completion Problems (QCP) to validate our

general approach

Page 11: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

11

SATzilla-07 Example Problem Problem distribution

QCP problems generated near phase transition [Gomes & Selman,

1997]

Solvers Eureka, OKsolver, Zchaff_Rand

Features Same as in previous work [Nudelman, et al. 2004]

Collect Data Compute instances’ features and determine solvers’ runtime

Pre-Solver & “Winner take all” Build Models Final solver selection

Page 12: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

12

Empirical Hardness Model (EHM) The Core of SATzilla --- EHM

Accurately predict algorithm’s runtime based on cheaply computable features

Linear basis function regression

Features () Runtime (y)

fw() = wT • 23.34

• 7.21

• …

• …

Page 13: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

13

Improve EHM (deal with censoring) Heavy-tailed behavior and censoring

Three ways for censored data Drop them Keep them as if finished at cutoff Censored sampling

Schmee & Hahn ‘s approach [1979] REPEAT

1. Estimate runtime conditional on EHM and real runtime bigger than cutoff runtime

2. Build new EHM with estimated runtimeUNTIL no more changes in EHM

Page 14: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

14

How to deal with censored data

A: Drop themB: Finished at cutoff

C: Censor sampling Censor point

Page 15: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

15

How to deal with censored data

A: Drop them

B: Finished at cutoffC: Censor sampling

Censor pointCensor point

Page 16: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

16

How to deal with censored data

A: Drop them

B: Finished at cutoff

C: Censor sampling Censor point

Censor point

Page 17: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

17

Improve EHM (using Hierarchal Hardness Models )

EHM often more accurate, much simpler when trained with SAT/UNSAT samples only [Nudelman, et al. 2004]

Building hierarchal hardness models by

approximate model selection Oracle Mixture of experts problem with fixed experts

Page 18: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla-07 for QCP

Average Runtime

Page 19: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla-07 for QCP

Empirical CDF

Page 20: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

2007 SAT Competition

Three submissions for 2007 SAT Competition

BIG_MIX for all three category (demo)RANDOM

HANDMADE

Page 21: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

21

SATzilla-07 for SAT Competition Target Distribution

Previous SAT competition and SAT Race Solver (with/without preprocessing, Hyper)

Eureka, Kcnfs2006, March_dl2004, Minisat2.0 Vallsat, Rsat, Zchaff_Rand

Features Reduce probing time to 1 second Only cheap features, total about 3 seconds

Pre-Solvers March_dl 5 seconds, SAPS 2 seconds

Page 22: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

22

SATzilla-07 for SAT Competition “ Winner take all” solver

March_dl2004 Final candidates

BIG_MIX Eureka, Kcnfs2006, March_dl2004, Rsat RANDOM March_dl2004, Kcnfs2006, Minisat2.0+ HANDMADE March_dl2004, Vallst, March_dl2004+,

Minisat2.0+, Zchaff_Random+

Page 23: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

23

SATzilla-07 for BIG_MIX

Page 24: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

24

SATzilla-07 for BIG_MIX

Pre-solvers

Feature time

Page 25: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

25

SATzilla-07 for RANDOM

Page 26: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

26

SATzilla-07 for RANDOM

Pre-solvers

Feature time

Page 27: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

27

SATzilla-07 for HANDMADE

Page 28: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

28

SATzilla-07 for HANDMADE

Pre-solvers

Feature time

Page 29: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

Conclusions

Page 30: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

30

Conclusions Can combine algorithms into portfolios,

improving performance and robustness SATzilla approach has been proven to be

successful in real world competition

With more training data and more solvers, SATzilla can be even better

Page 31: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

Ongoing research SATzilla for industrial category

Use the same approach, SATzilla is 25% faster and solves 5% more instances

Score function Optimize objective function other than runtime

Local search Improve SATzilla performance by using local

search solvers as component

Page 32: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

32

Special Thanks Creators of solvers

Alexander Nadel, Moran Gordon, Amit Palti and Ziyad Hanna (Eureka)

Marijn Heule, Hans van Maaren (March_dl2004) Niklas Eén, Niklas Sörensson (Minisat2.0) Oliver Kullmann (OKsolver) Knot Pipatsrisawat and Adnan Darwiche (Rsat 1.04) Daniel Vallstrom (Vallst) Yogesh S. Mahajan, Zhaohui Fu and Sharad Malik

(Zchaff_Rand)

Page 33: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla Pick for BIG_MIX

Page 34: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla Pick for RANDOM

Page 35: SATzilla-07: The Design and Analysis of an Algorithm Portfolio for SAT Lin Xu, Frank Hutter, Holger H. Hoos and Kevin Leyton-Brown University of British.

SATzilla Pick for HANDMADE


Recommended