approaches for adaptive choice-based conjoint...

19
Mixed Integer Programming (MIP) Approaches for Adaptive Choice-Based Conjoint Analysis Juan Pablo Vielma Massachusetts Institute of Technology Joint work with Denis Saure ISMS Marketing Science Conference, University of Southern California, Los Angeles, California. June, 2017.

Upload: truongphuc

Post on 01-Feb-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

MixedIntegerProgramming(MIP)ApproachesforAdaptiveChoice-BasedConjointAnalysis

JuanPabloVielmaMassachusettsInstituteofTechnology

JointworkwithDenisSaure

ISMSMarketingScienceConference,UniversityofSouthernCalifornia,LosAngeles,California.June,2017.

Page 2: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

Adaptive Choice-BasedConjointAnalysis

Feature SX530 RX100

Zoom 50x 3.6x

Prize $249.99 $399.99

Weight 15.68ounces 7.5ounces

Prefer � �

Feature TG-4 G9

Waterproof Yes No

Prize $249.99 $399.99

Weight 7.36lb 7.5lb

Prefer � �

Feature TG-4 Galaxy2

Waterproof Yes No

Prize $249.99 $399.99

Viewfinder Electronic Optical

Prefer � �

Werecommend:

Feature SX530 RX100

Zoom 50x 3.6x

Prize $249.99 $399.99

Weight 15.68lb 7.5lb

Prefer � �

• Today:Minimizevariance ofparameter estimates

Estimateofpreferenceparameter

1 /18

Page 3: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

ParametricModel= LogisticRegression

Feature Chewbacca BB-8

Wookiee Yes No

Droid No Yes

Blaster Yes No

Prefer? ☐ ☐

x

2x

1

0

@010

1

A = x

2

MNLRandomLinear Utility

Productprofile

n

Question:,

Uj = � · xj + ✏j

Xd

j=1�ix

ji

x

1 ⌫ x

2 , U1“�”U2

, � · z“�”0P�x

1 ⌫ x

2 | ��=

1

1 + e

��·z

z = x

1 � x

2

2 /18

Page 4: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

BayesianModelwithNormalPrior

Prior distribution

� ⇠ N(µ,⌃)

Answer likelihood

Posterior distribution

L (y | �, z) g (� | y, z )

g (� | y, z ) / � (� ; µ,⌃)L (y | �, z)

L (y | �, z) =�1 + e�y�·z��1

y = sign(� · z)

3 /18

Page 5: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

GeometricModels≈BayesianModel

�1

�2

�1

�2

�1

�2

Answer hyperplanePosterior set

Prior set

� 2 Pi � · z � 0 � 2 Pi+1

Method Response ErrorPolyhedralMethod(Toubiaetal.‘03,’04) NoProbabilisticPolyhedralMethod(T.etal.‘07) Yes,≈Bayesian

RobustMethod(BertsimasandO’Hair‘13) Yes,RobustEllipsoidalMethod(Saure andVielma‘16) Yes,=Bayesian

4 /18

Page 6: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

Bayesianv/sGeometric

Bayesian Geometric

ResponseError MNL None /Non-MNLor≈ MNL

Update IntegrationorMCMC

SimpleLinearAlgebra

QuestionSelection

Integration+Enumeration MIP

3/9/16, 12:08 PMAcademic Page of Juan Pablo Vielma

Page 3 of 3http://www.mit.edu/~jvielma/

Jennifer ChallisE62-571, 100 Main Street,Cambridge, MA 02142(617) 324-4378jchallis at mit dot edu

CollaboratorsShabbir Ahmed, Daniel Bienstock, Daniel Dadush, Sanjeeb Dash, Santanu S. Dey, Iain Dunning, RodolfoCarvajal, Luis A. Cisternas, Miguel Constantino, Daniel Espinoza, Alexandre S. Freire, Marcos Goycoolea,Oktay Günlük, Joey Huchette, Nathalie E. Jamett, Ahmet B. Keha, Mustafa R. Kılınç, Guido Lagos, MilesLubin, Sajad Modaresi, Sina Modaresi, Eduardo Moreno, Diego Morán, Alan T. Murray, George L.Nemhauser, Luis Rademacher, David M. Ryan, Denis Saure, Alejandro Toriello, Andres Weintraub, SercanYıldız, Tauhid Zaman

Affiliations

Links

Back to top© 2013 Juan Pablo Vielma | Last updated: 02/29/2016 00:47:25 | Based on a template design by AndreasViklund

3/9/16, 12:08 PMAcademic Page of Juan Pablo Vielma

Page 3 of 3http://www.mit.edu/~jvielma/

Jennifer ChallisE62-571, 100 Main Street,Cambridge, MA 02142(617) 324-4378jchallis at mit dot edu

CollaboratorsShabbir Ahmed, Daniel Bienstock, Daniel Dadush, Sanjeeb Dash, Santanu S. Dey, Iain Dunning, RodolfoCarvajal, Luis A. Cisternas, Miguel Constantino, Daniel Espinoza, Alexandre S. Freire, Marcos Goycoolea,Oktay Günlük, Joey Huchette, Nathalie E. Jamett, Ahmet B. Keha, Mustafa R. Kılınç, Guido Lagos, MilesLubin, Sajad Modaresi, Sina Modaresi, Eduardo Moreno, Diego Morán, Alan T. Murray, George L.Nemhauser, Luis Rademacher, David M. Ryan, Denis Saure, Alejandro Toriello, Andres Weintraub, SercanYıldız, Tauhid Zaman

Affiliations

Links

Back to top© 2013 Juan Pablo Vielma | Last updated: 02/29/2016 00:47:25 | Based on a template design by AndreasViklund

• EllipsoidalMethod:–MIP, andbridgestheGAP(QuestionSelection)

5 /18

Page 7: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

D-EfficiencyandExpected PosteriorVariance

cov(�) = ⌃1 cov(�) = ⌃2

y > 0y < 0

• ishardtoevaluate,non-convexandlargen

� ⇠ N(µ,⌃)

f(z, µ,⌃) := Ey,�

n

(det cov(� | y, z))1/mo

f(z, µ,⌃)maxz2{�1,0,1}n f(z, µ,⌃)

6 /18

Page 8: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

ReformulationfromV.andSaure ‘16

• D-efficiency=Non-convexfunction off(d, v)

Canevaluatewith1-dimintegral🙂

f(d, v)

LinearMIPformulation(standardlinearization)

PiecewiseLinearInterpolation

mean: d := µ · z

variance: v := z0 ·⌃ · z

f(z)

AlignswithselectioncriteriafromToubiaetal.’04:minimizemeanandmaximizevariance

�f(d, v)

7 /18

Page 9: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

EasytoBuildthrough&

• PiecewiseLinearOpt.jl (Huchette andV.2017)

5/27/17, 5:15 PMjoehuchette/PiecewiseLinearOpt.jl: Optimizing over piecewise linear functions

Page 2 of 2https://github.com/joehuchette/PiecewiseLinearOpt.jl

m=Model()

@variable(m,x)

To model the graph of a piecewise linear function f(x), take d as some set of breakpoints along the real line, and fd=[f(x)forxind] as the corresponding function values. You can model this function in JuMP using the following function:

z=piecewiselinear(m,x,d,fd)

@objective(m,Min,z)#minimizef(x)

For another example, think of a piecewise linear approximation for for the function $f(x,y) = exp(x+y)$:

usingJuMP,PiecewiseLinearOpt

m=Model()

@variable(m,x)

@variable(m,y)

z=piecewiselinear(m,x,y,0:0.1:1,0:0.1:1,(u,v)->exp(u+v))

@objective(m,Min,z)

Current support is limited to modeling the graph of a continuous piecewise linear function, either univariate or bivariate,with the goal of adding support for the epigraphs of lower semicontinuous piecewise linear functions.

Contact GitHub API Training Shop Blog About© 2017 GitHub, Inc. Terms Privacy Security Status Help

min exp(x+ y)

s.t.

x, y 2 [0, 1]

10

10

Automaticallyselect∆

Automaticallyconstructformulation(easilychosen)

3/9/16, 12:08 PMAcademic Page of Juan Pablo Vielma

Page 3 of 3http://www.mit.edu/~jvielma/

Jennifer ChallisE62-571, 100 Main Street,Cambridge, MA 02142(617) 324-4378jchallis at mit dot edu

CollaboratorsShabbir Ahmed, Daniel Bienstock, Daniel Dadush, Sanjeeb Dash, Santanu S. Dey, Iain Dunning, RodolfoCarvajal, Luis A. Cisternas, Miguel Constantino, Daniel Espinoza, Alexandre S. Freire, Marcos Goycoolea,Oktay Günlük, Joey Huchette, Nathalie E. Jamett, Ahmet B. Keha, Mustafa R. Kılınç, Guido Lagos, MilesLubin, Sajad Modaresi, Sina Modaresi, Eduardo Moreno, Diego Morán, Alan T. Murray, George L.Nemhauser, Luis Rademacher, David M. Ryan, Denis Saure, Alejandro Toriello, Andres Weintraub, SercanYıldız, Tauhid Zaman

Affiliations

Links

Back to top© 2013 Juan Pablo Vielma | Last updated: 02/29/2016 00:47:25 | Based on a template design by AndreasViklund

3/9/16, 12:08 PMAcademic Page of Juan Pablo Vielma

Page 3 of 3http://www.mit.edu/~jvielma/

Jennifer ChallisE62-571, 100 Main Street,Cambridge, MA 02142(617) 324-4378jchallis at mit dot edu

CollaboratorsShabbir Ahmed, Daniel Bienstock, Daniel Dadush, Sanjeeb Dash, Santanu S. Dey, Iain Dunning, RodolfoCarvajal, Luis A. Cisternas, Miguel Constantino, Daniel Espinoza, Alexandre S. Freire, Marcos Goycoolea,Oktay Günlük, Joey Huchette, Nathalie E. Jamett, Ahmet B. Keha, Mustafa R. Kılınç, Guido Lagos, MilesLubin, Sajad Modaresi, Sina Modaresi, Eduardo Moreno, Diego Morán, Alan T. Murray, George L.Nemhauser, Luis Rademacher, David M. Ryan, Denis Saure, Alejandro Toriello, Andres Weintraub, SercanYıldız, Tauhid Zaman

Affiliations

Links

Back to top© 2013 Juan Pablo Vielma | Last updated: 02/29/2016 00:47:25 | Based on a template design by AndreasViklund

8 /18

Page 10: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

MIP-basedAdaptiveQuestionnaires

Feature SX530 RX100

Zoom 50x 3.6x

Prize $249.99 $399.99

Weight 15.68ounces 7.5ounces

Prefer � �

Feature TG-4 G9

Waterproof Yes No

Prize $249.99 $399.99

Weight 7.36lb 7.5lb

Prefer � �

Feature TG-4 Galaxy2

Waterproof Yes No

Prize $249.99 $399.99

Viewfinder Electronic Optical

Prefer � �

Werecommend:

Feature SX530 RX100

Zoom 50x 3.6x

Prize $249.99 $399.99

Weight 15.68lb 7.5lb

Prefer � �

• Optimalone-steplook-aheadmoment-matchingapproximateBayesianapproach=EllipsoidalMethod

E�� |Y,X1, X2

cov

�� |Y,X1, X2

9 /18

Page 11: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

OptimalOne-StepLook-Ahead=MIP

Prior distribution

Feature SX530 RX100

Zoom 50x 3.6x

Prize $249.99 $399.99

Weight 15.68ounces 7.5ounces

Prefer � �

Feature TG-4 Galaxy2

Waterproof Yes No

Prize $249.99 $399.99

Viewfinder Electronic Optical

Prefer � �

Feature TG-4 Galaxy2

Waterproof Yes No

Prize $249.99 $399.99

Viewfinder Electronic Optical

Prefer � �

Feature TG-4 G9

Waterproof Yes No

Prize $249.99 $399.99

Weight 7.36lb 7.5lb

Prefer � �

minx

1,x

2f

�x

1, x

2�

• SolvewithMIPformulation

• Sampling:allprecomputed(2-dimgrid)

� ⇠ N�µi,⌃i

minx

1,x

2,d,v2Q

f(d, v)

10 /18

Page 12: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

Moment-MatchingApproximateBayesianUpdate

Prior distribution

Feature TG-4 Galaxy2

Waterproof Yes No

Prize $249.99 $399.99

Viewfinder Electronic Optical

Prefer � �

Answer likelihood

Posterior distribution

� ⇠ N�µi,⌃i

��

approx.⇠ N�µi+1,⌃i+1

• µ

i+1= E

�� | y, x1

, x

2�

• ⌃

i+1= cov

�� | y, x1

, x

2� • 1-dintegral:• Sampling:allprecomputed

I(d, v)

11 /18

Page 13: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

SimulationExperiments

• 16questions,2options,12features• SimulateMNLresponseswithknown𝛽*

• 100individual𝛽* sampledfromN(𝜇,𝛴)prior• Methods:– Polyhedral,Prob.Polyhedral,RobustandEllipsoidal– Allgetsameellipsoidalprior– All<30’’inter-question(exceptrobust<90’’)

• Metrics:– RMSEof𝛽 estimator,errorinmarketshareandD-eff.– Normalizedvalues=smallerbetter– Versions:Method,IndividualandHierarchicalBayesian– Sensitivity:Wrongprior𝜇,allerrorsinfirst/secondhalf 12 /18

Page 14: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

D-EfficiencyforIndividualBayesian

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

Baseline

WrongPrior SecondHalfErrors

FirstHalfErrors▲▲▲▲▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ●

◇◇◇◇◇◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeff2▲

▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

● ● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeff2

▲▲

▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeffdim

▲ ▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■■ ■

■■ ■ ■ ■ ■ ■ ■ ■ ■

● ●

●●

●● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeffdim

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■ ■

■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

●● ● ● ● ● ● ● ● ● ● ●

◇◇

◇◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeff2▲

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeff2

▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeffdim

▲▲

▲▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

● ●

●● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.5

1.5

HBDeffdim

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲▲▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■ ■

■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ●

◇◇◇

◇◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeff2▲

▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■

● ● ● ● ● ● ● ● ●

● ● ● ● ●

◇ ◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeff2

▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeffdim▲

▲▲ ▲

▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■ ■

■■■

■ ■ ■■■ ■ ■ ■ ■

●●

●●

● ● ● ● ● ●

● ● ● ● ●

◇ ◇◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeffdim

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ●

◇◇◇◇◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeff2▲

▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■

■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

● ● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

HBDeff2

▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

● ● ● ● ● ● ● ● ● ● ● ● ●

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.2

0.4

0.6

0.8

BDeffdim

▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

● ● ●

● ●●

●● ● ● ● ● ● ● ●

◇ ◇◇ ◇

◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.0

0.5

1.5

HBDeffdim

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

13 /18

Page 15: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

RMSEforMethodsEstimator

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

●●

●●

●●

●●

●● ●

● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.70

0.75

0.80

0.85

0.90

0.95

BRMSE

▲▲ ▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ●

●● ●

● ●●

● ● ●

● ● ● ● ●

◇◇

◇◇

◇ ◇ ◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.90

0.95

1.00

1.05

1.10HBRMSE

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲ ▲ ▲

▲ ▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■■ ■ ■

■■ ■ ■ ■

■ ■ ■ ■■

●●

●●

● ●●

●● ● ● ●

●● ● ● ●

◇◇

◇◇ ◇

◇◇

◇◇

◇◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.1

1.2

1.3

1.4

RMSE

▲▲

▲▲

▲▲

▲▲

▲ ▲▲

▲ ▲▲

▲■■

■■

■■

■■

■■

■■

■■

■■ ■

●●

●●

●●

●●

●●

●●

●●

◇◇

◇◇

◇◇

◇◇

◇◇

◇ ◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.95

1.00

1.05

1.10

BRMSE

▲ ▲

▲▲

▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●◇ ◇

◇◇

◇◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.2

1.4

1.6

HBRMSE

▲▲ ▲ ▲

▲ ▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■■ ■ ■

■■ ■ ■ ■

■ ■ ■ ■■

●●

●●

● ●●

●● ● ● ●

●● ● ● ●

◇◇

◇◇ ◇

◇◇

◇◇

◇◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.1

1.2

1.3

1.4

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲

▲▲

▲ ▲ ▲■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■

■ ■

● ● ● ● ● ● ● ● ●●

●●

●● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

◇ ◇◇

◇◇

◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE▲

▲ ▲▲

▲▲ ▲ ▲

▲▲

▲▲

▲▲ ▲ ▲

■■ ■ ■

■■

■ ■■ ■

■■

■■

■ ■ ■

●●

●●

●●

●●

●●

●●

◇◇ ◇

◇ ◇◇

◇ ◇ ◇◇

◇◇

◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.70

0.75

0.80

0.85

0.90

0.95

BRMSE

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■●

● ● ● ● ● ● ● ● ●

●●

●● ●◇ ◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.2

1.4

1.6

1.8

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲

▲▲

▲ ▲ ▲■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■

■ ■

● ● ● ● ● ● ● ● ●●

●●

●● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

◇ ◇◇

◇◇

◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

Baseline

WrongPrior SecondHalfErrors

FirstHalfErrors

14 /18

Page 16: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

RMSEforIndividualBayesianEstimator

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

Baseline

WrongPrior SecondHalfErrors

FirstHalfErrors

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

●●

●●

●●

●●

●● ●

● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.70

0.75

0.80

0.85

0.90

0.95

BRMSE

▲▲ ▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ●

●● ●

● ●●

● ● ●

● ● ● ● ●

◇◇

◇◇

◇ ◇ ◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.90

0.95

1.00

1.05

1.10HBRMSE

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲ ▲ ▲

▲ ▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■■ ■ ■

■■ ■ ■ ■

■ ■ ■ ■■

●●

●●

● ●●

●● ● ● ●

●● ● ● ●

◇◇

◇◇ ◇

◇◇

◇◇

◇◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.1

1.2

1.3

1.4

RMSE

▲▲

▲▲

▲▲

▲▲

▲ ▲▲

▲ ▲▲

▲■■

■■

■■

■■

■■

■■

■■

■■ ■

●●

●●

●●

●●

●●

●●

●●

◇◇

◇◇

◇◇

◇◇

◇◇

◇ ◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.95

1.00

1.05

1.10

BRMSE

▲ ▲

▲▲

▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ●

● ● ● ● ● ● ● ● ● ● ● ● ● ●◇ ◇

◇◇

◇◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.2

1.4

1.6

HBRMSE

▲▲ ▲ ▲

▲ ▲▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■■ ■ ■

■■ ■ ■ ■

■ ■ ■ ■■

●●

●●

● ●●

●● ● ● ●

●● ● ● ●

◇◇

◇◇ ◇

◇◇

◇◇

◇◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.1

1.2

1.3

1.4

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲

▲▲

▲ ▲ ▲■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■

■ ■

● ● ● ● ● ● ● ● ●●

●●

●● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

◇ ◇◇

◇◇

◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE▲

▲ ▲▲

▲▲ ▲ ▲

▲▲

▲▲

▲▲ ▲ ▲

■■ ■ ■

■■

■ ■■ ■

■■

■■

■ ■ ■

●●

●●

●●

●●

●●

●●

◇◇ ◇

◇ ◇◇

◇ ◇ ◇◇

◇◇

◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.70

0.75

0.80

0.85

0.90

0.95

BRMSE

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■●

● ● ● ● ● ● ● ● ●

●●

●● ●◇ ◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.2

1.4

1.6

1.8

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲▲

▲▲

▲ ▲ ▲■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■

■ ■

● ● ● ● ● ● ● ● ●●

●●

●● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

◇ ◇◇

◇◇

◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

15 /18

Page 17: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

RMSEforHierarchicalBayesianEstimator

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

Baseline

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●●

●●

●●

●●

●●

●● ●

● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.70

0.75

0.80

0.85

0.90

0.95

BRMSE

▲▲ ▲ ▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ●

●● ●

● ●●

● ● ●

● ● ● ● ●

◇◇

◇◇

◇ ◇ ◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.90

0.95

1.00

1.05

1.10HBRMSE

▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■■ ■ ■

● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.7

0.8

0.9

1.0

1.1

1.2

1.3

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral16 /18

Page 18: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

MarketshareforBaseline

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■

●● ● ●

●●

●●

●●

●●

● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

BRMSE

▲ ▲▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■

■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■● ● ●

● ●● ● ● ● ● ● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

1.0

1.5

2.0

2.5

3.0

3.5

HBRMSE

▲▲

▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲

■ ■ ■■

■ ■ ■■ ■ ■ ■

■ ■ ■■

■ ■●

● ● ● ● ●●

●●

●●

● ● ● ● ● ●

◇◇ ◇ ◇ ◇ ◇ ◇

◇◇ ◇

◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.9

1.0

1.1

1.2

1.3

1.4

1.5

RMSE

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲ ▲ ▲ ▲▲ ▲

▲ ▲ ▲ ▲▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■

● ●

● ●●

● ● ●●

● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.00

0.05

0.10

0.15

0.20

0.25

0.30share

▲▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲

▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■

■ ■ ■ ■ ■ ■■ ■ ■ ■ ■

●●

●●

● ● ●●

● ● ● ● ● ● ●

◇◇ ◇

◇ ◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.05

0.10

0.15

0.20

0.25

Bshare

▲ ▲

▲ ▲▲ ▲ ▲ ▲

▲▲ ▲ ▲ ▲

▲ ▲ ▲ ▲

■■

■■

■■

■ ■

■ ■ ■■ ■

■ ■

●●

● ●● ● ● ● ●

●● ● ● ●

◇ ◇◇

◇◇ ◇

◇◇

◇ ◇ ◇ ◇ ◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.10

0.15

0.20

0.25

HBshare

▲ ▲ ▲ ▲▲ ▲

▲ ▲ ▲ ▲▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■

● ●

● ●●

● ● ●●

● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.00

0.05

0.10

0.15

0.20

0.25

0.30share

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

▲ ▲ ▲ ▲▲ ▲

▲ ▲ ▲ ▲▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■

● ●

● ●●

● ● ●●

● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.00

0.05

0.10

0.15

0.20

0.25

0.30share

▲▲▲▲▲ ▲ ▲ ▲ ▲ ▲ ▲

▲ ▲ ▲ ▲ ▲ ▲

■■■■ ■

■ ■ ■ ■ ■ ■■ ■ ■ ■ ■

●●

●●

● ● ●●

● ● ● ● ● ● ●

◇◇ ◇

◇ ◇◇◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.05

0.10

0.15

0.20

0.25

Bshare

▲ ▲

▲ ▲▲ ▲ ▲ ▲

▲▲ ▲ ▲ ▲

▲ ▲ ▲ ▲

■■

■■

■■

■ ■

■ ■ ■■ ■

■ ■

●●

● ●● ● ● ● ●

●● ● ● ●

◇ ◇◇

◇◇ ◇

◇◇

◇ ◇ ◇ ◇ ◇◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.10

0.15

0.20

0.25

HBshare

▲ ▲ ▲ ▲▲ ▲

▲ ▲ ▲ ▲▲ ▲ ▲ ▲ ▲ ▲

■■ ■

■ ■■ ■ ■ ■ ■ ■ ■ ■

● ●

● ●●

● ● ●●

● ● ● ● ● ● ●

◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇ ◇

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

0.00

0.05

0.10

0.15

0.20

0.25

0.30share

▲ Probabilistic Polyhedral ■ Robust ● Bayesian Ellipsoid ◇ Polyhedral

Method IndividualBayesian

HierarchicalBayesian

17 /18

Page 19: Approaches for Adaptive Choice-Based Conjoint Analysisjvielma/presentations/ELLIPSOIDCONJOINT_MKS17.pdf · AdaptiveChoice-Based Conjoint Analysis Feature SX530 RX100 ... (Bertsimas

Summary

• MixedIntegerProgrammingforACBCA– n-variatefunctionto2-variatefunction+MIP– Precomputed2-variatePWLfunction– AdvancedMIPformulation+solver– Easytoaccesswith!• Askmeandgetasticker🙂

• Alsoforotherestimatorvariance/linearmodels• Significantlyfasterreductionofestimatorvariance–MaybetoofastforHB?

• Future:MIPflexibility➔ManagerialObjective

3/9/16, 12:08 PMAcademic Page of Juan Pablo Vielma

Page 3 of 3http://www.mit.edu/~jvielma/

Jennifer ChallisE62-571, 100 Main Street,Cambridge, MA 02142(617) 324-4378jchallis at mit dot edu

CollaboratorsShabbir Ahmed, Daniel Bienstock, Daniel Dadush, Sanjeeb Dash, Santanu S. Dey, Iain Dunning, RodolfoCarvajal, Luis A. Cisternas, Miguel Constantino, Daniel Espinoza, Alexandre S. Freire, Marcos Goycoolea,Oktay Günlük, Joey Huchette, Nathalie E. Jamett, Ahmet B. Keha, Mustafa R. Kılınç, Guido Lagos, MilesLubin, Sajad Modaresi, Sina Modaresi, Eduardo Moreno, Diego Morán, Alan T. Murray, George L.Nemhauser, Luis Rademacher, David M. Ryan, Denis Saure, Alejandro Toriello, Andres Weintraub, SercanYıldız, Tauhid Zaman

Affiliations

Links

Back to top© 2013 Juan Pablo Vielma | Last updated: 02/29/2016 00:47:25 | Based on a template design by AndreasViklund

18 /18