AVM f An Open-Source Implementation of the Alternating Variable - PowerPoint PPT Presentation

AVM f An Open-Source Implementation of the Alternating Variable Method Gregory M. Kapfhammer Phil McMinn SSBSE 2016 October 9, 2016

AVM is Everywhere Application Domains

AVM is Everywhere Application Domains Workloads

AVM is Everywhere Application Domains Testing Workloads

AVM is Everywhere Application Domains Testing Workloads SoftwareProductLines

AVM is Everywhere Application Domains Testing Workloads SoftwareProductLines AVM is used in varied domains

Exploring AVM Input Vector

Exploring AVM Input Vector � = ( x 1 , x 2 , . . . , x n ) x

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Objective Function

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Exploratory Moves

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Exploratory Moves Positive or negative “direction”?

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Pattern Moves

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Pattern Moves Improve objective value?

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Pattern Moves Improve objective value? Yes! pattern or No! explore

Exploring AVM Input Vector � � = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x Pattern Moves Improve objective value? Yes! pattern or No! explore

Exploring AVM Input Vector � � � = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x

Exploring AVM Input Vector � � � = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x Exploratoryand PatternMoves

Exploring AVM Input Vector � � � = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x Exploratoryand PatternMoves Consider all input vector variables

Exploring AVM Input Vector � � � � = = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x x Exploratoryand PatternMoves Consider all input vector variables

Exploring AVM Input Vector � � � � = = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x x Exploratoryand PatternMoves Revisit each x i in the input vector

Exploring AVM Input Vector � � � � = = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x x Exploratoryand PatternMoves Restart for local optimum

Exploring AVM Input Vector � � � � = = = = ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) ( x 1 , x 2 , . . . , x n ) x x x x Exploratoryand PatternMoves Continue until termination condition

AVM Innovations Search Algorithms

AVM Innovations Search Algorithms Geometric

AVM Innovations Search Algorithms Geometric Lattice

AVM Innovations Search Algorithms Geometric Lattice IteratedPattern

AVM Innovations Search Algorithms Geometric Lattice IteratedPattern Better search for many landscapes

AVM Innovations Search Algorithms Geometric Lattice IteratedPattern Provably faster for unimodal

AVM Innovations Representations

AVM Innovations Representations Decimals

AVM Innovations Representations Strings Decimals

AVM Innovations Representations Strings Decimals Integers

AVM Innovations Representations Strings Decimals Integers Handle real-world programs

Missing Features ✗ Data? ✗ Method?

Missing Features ✗ Data? ✗ Method?

Key Challenge Prior AVMs lack provably faster methods!

Tools Using AVM Test Generation

Tools Using AVM Test Generation AUSTIN

Tools Using AVM Test Generation AUSTIN EvoSuite

Tools Using AVM Test Generation AUSTIN EvoSuite SchemaAnalyst

Tools Using AVM Test Generation AUSTIN EvoSuite SchemaAnalyst AVM is used in many tools

Extracting AVM Fitness? Search?

Extracting AVM Fitness? Search?

Extracting AVM Fitness? Search?

Key Challenge Hard to extract AVM from custom software!

Rescued by AVM f

Rescued by AVM f OriginalAVMplusenhance- mentsfordataandsearch

Rescued by AVM f Clearimplementa- tionofcorealgorithms

Rescued by AVM f Adherestotheprinciples ofobject-orienteddesign

Rescued by AVM f Freeandopen-sourcesoft- warefrom avmframework.org

Rescued by AVM f

Rescued by AVM f NewApplicationDomain

Rescued by AVM f

Rescued by AVM f NewSearchAlgorithm

Design of AVM f

Design of AVM f Configure

Design of AVM f Configure Represent

Design of AVM f

Design of AVM f Objective

Design of AVM f Objective Report

Design of AVM f

Design of AVM f Seethepaperformoredesign andimplementationdetails

Design of AVM f Thetool’swebsitecontains extensivedocumentation

AVM f Demonstration

AVM f Demonstration javaorg.avmframework. examples.Quadratic

AVM f Demonstration javaorg.avmframework. examples.StringOptimization

AVM f Demonstration javaorg.avmframework. examples.GenerateInputData

AVM f Demonstration Input → Output StochasticBehavior

AVM f Demonstration Alreadyrun: gitclone&mvnpackage

AVM f ’s Contributions

AVM f ’s Contributions Overcomesthe challengesofusingAVM

AVM f ’s Contributions Provablyfastersearches andnewdatatypes

AVM f ’s Contributions Accessibleobject-oriented andalgorithmicdesign

AVM f ’s Contributions Open-sourcedownload from avmframework.org