Simplifying the design of workflows for large-scale data exploration and visualization

Simplifying the Design of Workflows

for Large- Scale

Data Exploration and Visualization

Juliana Freire Claudio Silva

http://www .cs. utah. edu/ "-J juliana http://www .cs. utah. edu/ "-J csilva

Workflows and Com putational Processes

representing and managing com plex com putations

- Simulations, data analysis, visualization, data integration

• They capture computation and analysis processes, enabling

- Automation, reproducibility, result sharing

• Workflows are rapidly replacing prim itive shell scr i pt s

- Apple's Mac OS X Autom ator, Microsoft Windows Workflow Foundation, and Yahoo! Pipes

• Business Workflows => Scientific Workflows

- Important differences!

Workflows: Scientific vs. Business

com puting intensive • Cater to a broad set of

users

• Ensure rules and

prescribed processes are followed

• Control flow (e.g., BPEL): State and side effects

c

Exploration and Workflows

• Workflows have been traditionally used to automate repet it ive tasks

• I n exploratory tasks, change is the norm!

Data analysis and exploration are iterative processes

Data Data ~ Workflow Specification .... ~ Data Product Manipulation

Microsoft eScience Workshop, 2008

T ~ Perception & Cognition Exploration .... ~ Knowledge User

Figure modified from J. van Wijk, IEEE Vis 2005

•

Data Exploration and Workflows

raw data: CT scan

Files (workflow specifications) anon4877 _voxel_scale_1_zspace_20060331.srn anon4877 _textureshading_20060331.srn anon4877 _textureshading_planeO_20060331.srn anon4877 _goodxferfunction_20060331.srn Notes r lI\..~t~til L V~SUIA tLZlAtLDIA ,

WI.: Aototeot textuye til V1 Aototeot pLtilll\..e to

VI . ,

Explorat ion and Creat ivity Su pport

• Exploratory processes require reflective reasoning

• "Reflective reasoning requires the ability to store temporary results, to make inferences from stored knowledge, and to follow chains of reasoning

backward and forward, sometimes backtracking when a prom ising line of thought proves to be unfruitful .... the process is slow and laborious"

Donald A. Norm an

• Need external aids-tools to facilitate this process

- Creativity support tools [Shneiderman, CACM 2002]

• Need aid from people-collaboration

Data Exploration and Workflows: Issues

• Hard to assem ble and iteratively refine workflows

• Combine many tools and libraries: Need in-depth knowledge to weave them together

• No support for reflective reasoning

- E.g., history of the exploration trail maintained manually through file-nam ing conventions and detailed notes

- Hard to understand the exploratory process and relationships am ong workflows

• Lack of support for collaboration

Existing systems fail to provide the necessary

infrastructure for exploratory tasks. As a result, the

generation and maintenance of workflows is a major

VisTrails: Managing Scientific Exploration

• Goal: reduce time to insight

• Build infrastructure to streamline exploratory tasks

such as data analysis and visualization

• Support for collaboration

• Usability-provide tools and intuitive interfaces

• The VisTrails System: an open-source provenance-enabled scientific workflow system

- > 6,000 downloads since 2007

- Used in many applications: environmental modeling (OHSU), physics sim ulation (Cornell, LANL), medical studies (University of Utah), ...

Outline

• Using provenance to support reflective reasoning

• Explori ng and re- u si ng proven ance

- Querying workflows by exam pie - Creating workflows by analogy - Auto-com pletion for workflows

• Emerging applications

c

Change-Based Provenance

• Captures provenance of workflow

evolution

• Records user actions

• Provenance

=

com putational tasks

- Add a module, add a connection, change a parameter value

addModule deleteConnection addConnection addConnection setParameter

Q

~n;V

-E~~

•

Change-Based Provenance

• Records user actions

• Provenance

=

- Add a module, add a connection, change a parameter value

• Extensible change algebra

• A vistrail node v_t corresponds to the workflow that is constructed by the sequence of actions from the root to v_t

v_t

=

xn

xn_

1 ° ... 0

x

1 0

0

[Freire et ai, I PAW 2006]

Microsoft eScience Workshop, 2008

vistrail

Exploring the Change Space

• Script i ng workflows: Param eter explorat ions are

sim pie to specify and apply

• Exploration of param eter space for a workflow

v

(setParameter(id_n, value_n) 0 • • • 0 (setParameter(id_{t ,} value _{t )} 0 v

t )

l .... lVi.TraiI>Oulidor lorminalor,lIml g~~

"lkCan..era Set'/\eWUP(O,O, ,I) set9osIIon(7~S. ~S3, 369) 5«Foc-<13S. 135, ISO) vtkContourfilt ... 5et-.iue(O.67) vtkStructu,edPointsRe_

Exploring the Change Space

• Script i ng workflows: Param eter explorat ions are sim pie to specify and apply

• Exploration of param eter space for a workflow

v

(setParameter(id_{n ,} value_n) 0 • • • 0 (setParameter(id_{t ,} value_{t )} 0 v

t )

• Exploration of multiple workflow specifications

(addModule(id;, ..

J

1 )

(addModule(id;, ..

J

• Results can be conveniently compared in the VisTrails spreadsheet

• Can create an im at ions too!

• Caching to avoid redundant com putations [Bavoil et aI., I EEE Vis 2005]

Com puting Workflow Differences

• No need to com pute subgraph

isom orph ism!

• A vistrail is a rooted tree: all

nodes have a com m on ancestor-diffs are

well-defined and simple to

compute

vt1

=

0

vt

=

0

vt ₁-vt₂ = _{{Xi' X i- t ' ... , X t ,} 0)

( X_{j ,} x_{j - 1, ••} .,X 1 , 0)

• Different sem ant ics:

Collaborative Exploration

• Collaboration is key to data exploration

- Translational, integrative approaches to science

• Store provenance information in a database

• Synchronize concurrent updates through locking

- Real-tim e collaboration [Ellkvist et aI., I PAW 2008]

• Asynchronous access: similar to version control

system s

- Check out, work offline, synchronize - Users exchange patches

• Synchronization is sim pie-provenance is

monotonic

• No need for a central repository-support for

distributed collaboration

- For details see Callahan et ai, SCI I nstitute Technical Report, No. UUSCI-2006-016 2006

Change- Based Provenance: Su m mary

• General: Works with any system that has undo/redo!

• Concise representation

• Uniform Iy captures data and workflow provenance

- Data provenance: where does a specific data product come from?

- Workflow evolution: how has workflow structure changed over time?

• Results can be reproduced

• Detailed information about the exploration process

• Provenance beyond reproducibility:

- Support for reflective reasoning

- Scalable exploration of the par am eter space-results can be com pared side-by-side in the spreadsheet

• Storing detailed information is important, but not

enough!

• Need appropriate user

interface and operations to leverage information

- Understand and re- use the history

• Simplify the creation of new workflows

Looking for Exam pies

• Need to query workflow collection:

- Find workflows that process a particular type of file - Find workflows that output a particular data product

- Find workflows that contain a given module or sequence of modules

• Workflow are graphs: hard to specify queries using text

Querying Workflows by Exam pie

• WYSIWYQ --

What You See Is

What You Query

• I nterface to create workflow is sam e as to query

Pipeline Interface 000000 vtkStructuredPolntsReader o 00000000 vtkOpenGLVolumeTextureMapper3D

Microsoft eScience Workshop, 2008

o

[Scheidegger et al., TVCG 2007

Query Result

.,-;..

---Query Interface

Refining Workflows

• Com plex workflows are hard to create

- Dam ain knowledge

- Fam iliarity with different tools Steep learning curve

r

... T ... ...

Bit fl:Iit »I'll' &M1 ~~ ~

JW ~ ~ PI' 0 ~Imt\l"'ton .... lQof:'Y p ... ~ .. r:~a!lon

:~DJ-·~- i~:

--

:0".

~'

.

Refining Workflows by Analogy

• Leverage the wisdom of the crowds in shared

provenance

- Sam e workflow refinem ents are com man, e. g., change the rendering technique, publish im age on the Web

• Apply refinem ents by analogy, autom atically

[Scheidegger et ai, IEEE TVCG 2007]

AnalogyTemplate

MicrosoTt e~clence vvorKsnop, ~UUts

~

.

Automatically constructed visualizations

o

Refining Workflows by Analogy

poe Report

ProteIn Title

Authors

Atom Count

is to

NEURAL CELL ADHESION MOLECULE, MODULE 2, NMR, 20 STRUCTURES P.H.JENSEN, V.SOROKA, N.K.THOMSEN, V.BEREZIN, E.BOCK, F.M.POULSEN C:9560 H: 15440 N:2580 0: 2680 5:60

Links POB Entry

as

7

o

The Analogy Algorithm

1. Com pute difference: 8(A,8)

- Just like a patch! - But ...

D = 8(A,B) 0 C may not be a valid

workflow

2. Find correspondences between A

and C: map(A,C)

Diffuse similarity scores across the

product graph Axe using Eigenvalue

is to

as

lIJ

D

111 = DIFF(A, 8) 1 ~.

A

decom pos ition s PDBReport

~~---~~~temT~me-=STRU~CTU~REO~FTH~E ~

3. Com pute mapped difference Authors ~~~!f!~f~~f~~~:~~TER

1\ (A 8) (A C) 1\( A 8) Atom Count C:5934

UAC '

=

4. D

=

C

Microsoft eScience Workshop, 2008

C: 20 N: 12 NA: 4 0:44 P: 6 Links .E.Qa..E.o1[y. ~ ________ ~ ______ ~ D

The Analogy Operation

• Allows workflows to be refined without requiring users to direct Iy m od ify the speeifieat io n

• Basis for scalable updates

• Analogies are not foolproof

- If it works, great. If it doesn't, it may help - User can edit and fix the new version

• I m prove by

- Using domain knowledge

The Need

for

Guidance

Workflow Design

In

vtkChacoReader 00000000 000000000 OOOOOOOO[ 0 DO

vtkCh, 000000 lateErode31

1 vtkWarpVector vtkExtractEdgE vtkFieldDat; vtklmplicitSum n

vtkCh. vtkPlecewlseFunction DO

~~rD 0 J - ouuuuuu __ :.JUUUUL 0000000000 JOO 00000000

vtk vtkRungeKutta2 IUnaY;jl)OO :one vtkLODActor vtkAxes vtkTransformFilter metl'1 vtkSphereSource

vtk 0 0 DO . DO

vtk L-JL-JL-J nn~ooo ~UUUUUOOOOOOOO

D~D[tmlc..tL"J'- tklmageActor

I

o vtkCylin vtkProperty2D - - 0

vtkTransform 01000000 0 0000000000 ...JL-JL-JL-JL-JL-JL-JL-JL-JL-JL-JL-J&:::J

DO I ... JrowSource vtkPolyDataMapper vtklmageReslice

VTKC~~tTnCnnr.rAtA r'r 0 DO

~kD'Dc10nOOOOI 0 IctorStyle 00000000 0_______ Juuu ...

vt vtkElevation vtkCylindricalTransform -.JUUUL 0 vtkDICOMlmageRe vtkLlneSourc~r:J !IIDataToPointDat~n

~ ~LJLJLJLJLJLJLJLJ 0 vtkActor DO JOOOOOOOOOO

vtkConeSource vtkBVUReader ""] 0 0 0 0 0 0 0 0 vtkActor2D 0 vtkGlyph3D 000

vtkConnectivityFilter vtkP rt I vtklmageFFT vtkTextActor

~~~~IODOOOOOOOOOOO ~ rope y 0 JGL-JL-JL-J oog~ooooo 0 0

vtkCo vtklmageResample b Ax LJ

A t 2D vtkHyperOctree vtkWarpScalar

vtkCo DO e es c or - vtkParametricSuperEllipsoid 0 L

vt~ 00000000 [

00-vtkCoordinate 0000______ S bd' .. F'lt VII\\,onnec;;t uuuuuuuu

)erOctrE vtkLoop u IVlslon I er

0000000 0 DO vtkCamer;l vtklmageDataGeometryFilter vtkStructuredGridReader 00000000 DO DO :J 0 0 0 0 0 0 0 0 vtkM-000000000 . edGraalem~naaer- 000000000 vtkCursor2[ WWWWWWWW vtkDelaunay2D c . . [ vtkXVZMolReader vtkCursor3[ vtklmageClip~~ vtkWelghtedTranSformFllteor 0 0000000 DO vtkCurvatun vtkCutter 00000000000 tkFrustumCoverage\,ul u u u u u u u u u u u u t..J 0 0 0 0 0 0 0 0 0 U U U

vtkCylinder . .&1,"1 n'T'''''U~eader vtkMergeFllter vtkRungeKutta4

V~CYI~n~e.rsource 00000000 DO. nnnnnoooo 0

~ , vtkPla 00000000 rammableF 00000000 0000000

00000 vtkButterflySubdivisionFilter 100000 vtkSurfaceReconstructionFilter vtkRenderer

vtkScalarE DO DO 0

u VlK:")ueamTracer

DO

VisCom plete: A Workflow

Recom m endation System

• Mine provenance collection: Identify graph

fragments that co-occur in a collection of workflows

• Predict sets of likely workflow additions to a given

part ial workflow

• Similar to a Web browser suggesting URL

com pletions ~kOSPFiIt'rGroup \<tkOataOb,ectlbDataSlt .. IAkOttaS.I'SutfaceFdtlt ~kOatiiSeIToDataObl·ct ... -.tkOataSelltlanoleFlhv oA.kOecunal.pO~ln.F,'t.r \rtkElevatlonFllter IAkFleldDiltaTDAttribute. IAkGenerlcContoutFll!.er 'A.kGenerICGeomttryRlter-\AkGenerlcG/yptlJDAlter IotkGlnelicOUtllMRlttr \otkGener\cProbeFlller \OtkGtometf)flll:er IAkGradlentFiller IAkG~phLayOUtl1lt.r tA.kH~chlc.lOatilLewt., tAkH\et.rchicalDataS.tG.

[Keep et al., IEEE Vis 2008]

Builder

.BI. ~dlt lilI-am ~ ElIlr.n Help

1I1kDataSeITo08IaOb,eCI ... \o1kDalaSft"',an~F"" VlI:Declm.teP~nef1lter ..n.kElevauonAlttr vtkFI.ldO.t.ToAttnbut .... vtkGtnlncContourFllttf vllffia"aNCGoomltl)'flhaf vtkGenlncGlyphlOFliter vtkG"n.rkOut~n"FNt.r vtkGenencProbeFllter IItkGtOrMll)fiIttr \/Iir:GJadl.ntAIt,t \'tkG~hl.¥lIlRIt.,. vt"HlerarchlcaIOalaLew:l •• vtkHler.rchlc .. fO .. tiiSetG ... vtkIottP8fOdr . . Contourf •••

VisCom plete: A Workflow

Recom m endation System

• Identify graph fragments that co-occur in a

collect ion of workflows

• Predict sets of likely workflow additions to a given

part ial workflow

• Similar to a Web browser suggesting URL

com pletions

OOOOOD

Microsoft eScience Workshop, 2008

o 0000

vtkOataSetRaadar

D

VisCom plete: Dem

http://www.cs.utah.edu/''-Jjuliana/videos/viscomplete_h_264.mov

[Koop et al., IEEE Vis2008]

Resu Its Su m mary

• Eliminates over 50% of actions

• Selected completions are almost always in the first

four suggestions

• A database of sim pie pipelines can aid users

constructing more complex pipelines

• See [Koop et al., TVCG 2008] for details on how the

path database is constructed and on the completion algorithm

Conclusions and Future Work

• Appropriate support for exploratory tasks is

essential for a wider adoption and more effective use of scientific workflow systems

• Provenance can be used to support reflective

.

reasoning

• Intuitive interfaces for simplifying the construction and refinem ent of workflows

• Sharing workflows provenance at a large scale creates new opportunities

- Workflow/ provenance repositories; provenance- enabled publications

- Scientists can learn by exam pie; expedite their scientific

Acknowledgments: Funding

• This work is partially supported by the National

Science Foundation, the Department of Energy, an IBM Facu Ity Award, and a Un iversity of Utah Seed Grant.

Microsoft eScience Workshop, 2008

DOE SciDAC VISUALIZATION AND ANALYTICS CENTER FOR ENABLING TECHNOLOGIES

1 • •

1

.

1

.

-- -- -- --_{--- ------ - -}

---- --

---

Acknowledgments: People

For more info about VisTrails

Visit: http://www.vistrails.org