Online survey for collective clustering of computer generated architectural floor plans

Page 1

Onlinesurveyforcollectiveclusteringof computergeneratedarchitecturalfloorplans

DavidSousa-Rodrigues⇤ 1 ,MafaldaTeixeiradeSampayo2 , EugénioRodrigues3 ,AdélioRodriguesGaspar4 ,ÁlvaroGomes5 , andCarlosHenggelerAntunes5

1 CentreofComplexityandDesign,FacultyofMaths,ComputingandTechnology, TheOpenUniversity,MiltonKeynes,UnitedKingdom

2 CIES,DepartmentofArchitecture, LisbonUniversityInstitute,Lisbon,Portugal

3 ADAI,LAETA,INESCCoimbra,DepartmentofMechanicalEngineering, UniversityofCoimbra,Coimbra,Portugal

4 ADAI,LAETA,DepartmentofMechanicalEngineering; UniversityofCoimbra,Coimbra,Portugal

5 INESCCoimbra,DepartmentofElectricalandComputerEngineering, UniversityofCoimbra,Coimbra,Portugal

ExtendedAbstract

Keywords: OnlineSurvey,GenerativeDesign,Clustering,CollectiveIntelligence,Floor PlanDesign,Architecture,Education.

Theaimofthisstudyistounderstandwhatarethecollectiveactionsofarchitecture practitionerswhengroupingfloorplandesigns.Theunderstandingofhowprofessionalsandstudentssolvethiscomplexproblemmayhelptodevelopspecificprogrammes fortheteachingofarchitecture.Inaddition,thefindingsofthisstudycanhelpin thedevelopmentofquerymechanismsfordatabaseretrievaloffloorplansandthe implementationofclusteringmechanismstoaggregatefloorplansresultingfromgenerativedesignmethods.Thestudyaimstocapturehowpractitionersdefinesimilarity betweenfloorplansfromapoolofavailabledesigns.Ahybridevolutionarystrategy isused,whichtakesintoaccountthebuilding’sfunctionalprogramtogeneratealternativefloorplandesigns[1–3].Thefirststepofthismethodologyconsistedinanonlinesurveytogatherinformationonhowtherespondentswouldperformaclustering

Correspondingauthor: david.rodrigues@open.ac.uk (DavidSousa-Rodrigues)

ExtendedabstractacceptedforICTPI’15conference,June17–19,MiltonKeynes,UnitedKingdom–http://www.ictpi15.info/

1

task.Onlinesurveyshavebeenusedinseveralapplicationsandareamethodofdata collectionthatconveysseveraladvantages.Whenproperlydevelopedandimplemented,asurveyportraysthecharacteristicsoflargegroupsofrespondentsonaspecific topicandallowsassessingitsrepresentation.Severaltypesofsurveysareavailable; e.g.questionnaireandinterviewformats,phonesurvey,andonlinesurveys,whichcan becoupledwithinferenceenginesthatactanddirectthesurveyaccordingtorespondents’answers[4,5].Inthepresentstudy,thesurveywasposedasanonlineexercisein whichrespondentshadtoperformapre-definedtask,whichmakesitsimilartorunninganexperimentinanonlineenvironment.Theexperimentaimedtounderstand theperceptionandcriteriaofthetargetpopulationtoperformtheclusteringtaskby comparingtheresultswiththerespondents’answerstoaquestionnairepresentedat theendoftheexercise.

Figure1:Agedistributionofrespondents

Thetargetgroupofthissurveyisindividualswhosedailyactivitiesarerelatedto architecture,i.e.architects,architecturestudents,civilengineers,andurbanplanners. ThepoolofparticipantsinhabitsmainlyinPortugalandtheagesrangebetween18and 50yearsold.Figure1depictstheagedistributionoftherespondents.

Thetaskwasperformedonlinethroughawebapplication.Fromapopulationof 72floorplans,twelverandomlyselecteddesignsarechosenanddisplayedonscreen. Theuseristhenaskedtodrag-and-droptoaspecificscreenareatheplansthathe considerssimilar(seefigure2).The72floorplansweregeneratedusingtheEvolutionaryProgramfortheSpaceAllocationProgram(EPSAP)[1–3].Thisalgorithmis capableofproducingalternativefloorplansaccordingtothesameuser’spreferences andrequirementssetasthefunctionalprogram.Thisdefinesthetypeofbuildingto

2

Figure2:TaskPanelofthesurvey.Usersmustdrag-and-droptotheblueareathefloor planspresentedontheleftaccordingtotheirnotionofsimilarity.

begeneratedandthedesignconstraints.Thesolutionsgeneratedwereasingle-family housewiththreebedrooms,onehall,onekitchen,alivingroom,onecorridorandtwo bathrooms.Abathroomandallthebedroomsareconnectedtothecorridorandall remainingspacesareconnectedtothehall.Thekitchenalsopresentsaninternaldoor connectingittothelivingroom.Oneofthebathroomsservesthepublicareasofthe housewhiletheotherconnectstothecorridoroftheprivateareaofthehouse.All innerroomshavedoorsof90cmwidth,theexceptionbeingthelivingroomdoorsthat are140cm.Withtheexceptionofthecirculationareasandoneofthebathrooms,all areashaveatleastonewindow—thelivingroomhastwo.Thehallhasadoortothe exteriorfacingNorth.Nootherrestrictionswereimposedonthefunctionalprogram ofthisproject.Allsolutionspresentedtotheparticipantswerepreviouslygenerated andtherewasnohumaninterventionintheirselectionforthisexercise.Theparticipantswereaskedtoperformaniteratedtask—tentimes—ofselectingsimilarfloor plans.Attheendofthoseteniterations,afinalformispresentedfortherespondentto identifythecriteriausedintheselectionofthedesigns.Atthismomenttheparticipant couldalsoreview—butnotchange—hispreviousselections.Aftersubmissionthe exercisewasfinished.Thedataobtainedwereanalysedaftertheconstructionoftwo squarematrices—onerepresentingineachentrythenumberofco-visualisationofthe floorplans,i.e.thenumberoftimesfloorplaniandfloorplanjwereshowninthesame iteration;andthesecondmatrixrepresentingthenumberofco-selectionsofthefloor plansbytheuser,i.e.thenumberoftimesthepairwasselectedassimilar.Thefirst matrixistheco-occurrencematrixwhilethesecondistheco-selectionmatrix.Anorm-

3

alisedmatrixisconstructedbydivisionofthetwopreviousmatrices.Thenormalised matrixgivesthefractionoftimeseachpairoffloorplanswasselected.Thismatrixcan beunderstoodasanadjacencymatrixwheretheentriesrepresenttheweightsofthe connectionsbetweentwofloorplandesigns.Theresultspresentsomebackgrounduncertaintyanditisnecessarytodefineaminimumthresholdfortheentriesofthematrix. Thevalueofthethresholdisvariedtoidentifythestructureoftheselectionprocess. Theresultingfloorplan’snetworkrepresentsthestructureoftheselectionmadebythe participants.Thisnetwork—undirectedandweighted—ispartitionedwiththeedge betweennesscommunitydetectionalgorithmbyGirvanandNewman[6].Thisisadivisivehierarchicalalgorithmthataimstofindcommunitiesbymaximizingthevalue ofmodularity—networkswithhighmodularityhavedenseintra–clusterconnections butsparseconnectionsbetweenverticesofdi↵erentclusters.Thegraphandtheresultingpartitionischaracterisedaccordingtodiverseproperties—degreedistribution, clusteringcoecient,assortativity,small-world,andscaleinvariance.

Figure3:Communitiesdetectedforthefloorplansdesignswiththresholdof15%

Weshowhowtopologicalpropertiesemergeinthefloorplan’snetwork,andcharacteriseitbyshowinghowthecommunitiesareidentifiedbythecollectiveanswers oftherespondents.Inthecasewhennothresholdisappliedtotheadjacencymatrix theresultingnetworkpresentsasinglegiantcomponentwith15completecliques— subsetsofverticeswheretheinducedsubgraphiscomplete,i.e.everytwoverticesare connected—andanetworkdiameterof2.Whenapplyinga15Byperformingasweep ofthethresholdoftheminimumpercentageofselections,whentwoplansareshownin common,itispossibletoidentifythefloorplansthataretherootsofthedi↵erenttypologies.Thesetsoffloorplansarenotdefinedinahierarchicalmannerbutsomepairs offloorplanswillnaturallybeco-selectedmoreoftenthanothers.Thushierarchiesof

4

pairsoffloorplansbasedaredefinedontheirco-selectionfrequency.Theunderstandingofhowpeopleperformcertaintasksiscrucialforthedevelopmentofeducation strategiesfocusedonimprovinglearningatuniversityleveleducation.Severalstudieshavebeenproposedthatincludetheparticipationofthecrowdandarebottomup learningprocesses,e.g.peerassessment[7]wherestudentsmarkeachother’swork.In thisstudytheprocessofgroupingfloorplansisinvestigatedtounderstandthecriteria usedbythestudentsandotherpractitioners.Theresultsarepresentedanddiscussed inlightofteachingstrategiesforthearchitectureeducationattheuniversitylevel. Theresultsshowhowcollectiveactiononsimpletaskscanleadtotheemergenceofthe solutionforthecomplextaskofdefininghierarchiesofsimilarityinfloorplan’sdesigns andidentifyingthecriteriausedbyaclassofprofessionals.Theresultsobtainedinthis workareimportantforfuturedevelopmentofICT-mediatedstrategiesforarchitecture educationandprofessionalpractitioners.Theywillalsoimpactotherapplicationssuch asfloorplandesigndatabaseretrievalandaggregationofsimilarsolutionsthatresult fromgenerativedesignmethods.Thecriteriareportedbytherespondentsvariedand canbeincorporatedinmachinelearningalgorithmstoperformtheclusteringtasks presentedtohumansinwaysthatmimicexperts’actions.

References

[1]RodriguesE,GasparA,GomesÁ.Anevolutionarystrategyenhancedwithalocal searchtechniqueforthespaceallocationprobleminarchitecture,Part1:Methodology.ComputerAided-Design.2013;45(5):887–897.

[2]RodriguesE,GasparA,GomesÁ.Anevolutionarystrategyenhancedwithalocal searchtechniqueforthespaceallocationprobleminarchitecture,Part2:Validation andPerformanceTests.ComputerAided-Design.2013;45(5):898–910.

[3]RodriguesE,GasparA,GomesÁ.Anapproachtothemulti-levelspaceallocation probleminarchitectureusingahybridevolutionarytechnique.AutomationinConstruction.2013November;35:482–498.

[4]UrbanoP,Sousa-RodriguesD.RuleBasedSystemsAppliedToOnlineSurveys.In: IADISWWW/InternetConference.Freiburg;2008.

[5]UrbanoP,Sousa-RodriguesD.TheAdvantageOfUsingRulesinOnlineSurveys. RevistadeCiênciasdaComputação.2008;III(3).

[6]GirvanM,NewmanMEJ.Communitystructureinsocialandbiologicalnetworks. ProceedingsoftheNationalAcademyofSciences.2002;99(12):7821–7826.

[7]deSampayoMT,Sousa-RodriguesD,Jimenez-RomeroC,JohnsonJH.PeerAssessmentinArchitectureEducation.In:InternationalConferenceonTechnologyand Innovation.Brno,CzechRepublic;2014.

5

OpenResearchOnline

TheOpenUniversity’srepositoryofresearchpublications

Onlinesurveyforcollectiveclusteringofcomputergeneratedarchitecturalfloorplans

ConferenceItem

Howtocite: Rodrigues,David;TeixeiradeSampayo,Mafalda;Rodrigues,Eugo;RodriguesGaspar,Ado;Gomes, lvaroandHenggelerAntunes,Carlos(2015).Onlinesurveyforcollectiveclusteringofcomputergenerated architecturalfloorplans.In:15thInternationalConferenceonTechnology,PolicyandInnovation,17-19June 2015,TheOpenUniversity,MiltonKeynes(forthcoming).

ForguidanceoncitationsseeFAQs

c 2015TheAuthors

Version:Proof

CopyrightandMoralRightsforthearticlesonthissiteareretainedbytheindividualauthorsand/orothercopyrightowners.FormoreinformationonOpenResearchOnline’sdatapolicy onreuseofmaterialspleaseconsult thepoliciespage.

andotherresearchoutputs
oro.open.ac.uk

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.
Online survey for collective clustering of computer generated architectural floor plans by Mafalda Teixeira de Sampayo - Issuu