SlideShare a Scribd company logo
1 of 14
Download to read offline
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
A discussion on the performance of the
Coordinate Exchange Algorithm
Tian Tian
Department of Mathematics, Statistics and Computer
Science
University of Illinois at Chicago
09 February 2016
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Background of DCEs
Background
A product/service of interest
K attributes each with lk levels
L = K
k=1 lk alternatives/profiles
Each choice set consists of J profiles
Q = L
J choice sets
Each design consists of S choice sets
M = Q
S candidate designs (design space)
Goal
–to estimate the appeal of each attribute
–to predict the behaviors of consumers
–to emulate market decisions and forecast market demands
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Background of DCEs
Two critical components
Parameter setup in the probabilistic model
(we model the probability a consumer chooses profile j in choice
set s — pjs)
An efficient algorithm that finds the optimal discrete choice
design(s)
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Challenges and motivation
Challenges
First challenge:
The probabilistic choice models are nonlinear ⇒ efficiency of the
design depends on the unknown parameters
Possible solutions:
Linear optimal designs where parameters are set to zeros
Locally optimal designs where parameters are assigned with
nonzero values
Bayesian optimal designs where parameters are introduced with
a prior distribution
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Challenges and motivation
Challenges (Cont.)
Second challenge:
Difficult (impossible) to theoretically construct locally/Bayesian op-
timal designs under a nonlinear multi-dimensional probabilistic choice
model.
Existing algorithms:
The mFA (modified Federov algorithm)
The RSA (relabeling and swapping algorithm)
The RSCA (relabeling, swapping, and cycling algorithm)
The quadrature scheme
The CEA (coordinate-exchange algorithm)
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Challenges and motivation
Motivation
Question:
Is the resulting design generated from the CEA indeed optimal, or
at lease highly efficient?
An example:
3 × 3 × 2/3/8 ⇒ (3×3×2
3 )
8
> 4 × 1018
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Review of the MNL model
The MNL model
The MNL (multinormial logit) probability pjs, j = 1, ..., J, s =
1, ..., S, is modeled as follows,
pjs =
exp(xjsβ)
J
i=1
exp(xisβ)
– ˜K = K
k=1(lk −1) is the total number of main effect levels subject
to the ”zero-sum” constraint
– xjs is a ˜K × 1 vector denoting the attribute levels of profile j in
the sth choice set
– β is a ˜K × 1 vector of parameter representing the main effects of
the attribute levels
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Review of the CEA
The CEA
1 Start with a randomly chosen starting design
2 Exchange levels for each attribute in each profile in each choice
set in the design
3 A level is updated when the new design results in a smaller
criterion value
4 A complete cycle of iteration terminates when the algorithm has
reached the last attribute in the last profile in the last choice
set
5 The algorithm goes back to the very first attribute and contin-
ues another round
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Review of the CEA
The CEA (Cont.)
6 A complete try of iteration terminates when no substitution is
made in a whole cycle or other stopping rule is met
7 To avoid poorly local optima, T tries of the algorithm are re-
peated each with a different starting design
8 The optimal design is chosen from the T resulting designs.
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Discrete and approximate design
To take advantage of the structure under the continuous/approxi-
mate design framework.
Approximate/continuous: ˜ξ = {(Cq, wq)|q = 1, ..., Q}.
Discrete/exact: ξ = {Cs|s = 1, ..., S} = {(Cq, wq)|q = 1, ..., Q}
with wi = 1/S for i = 1, ..., S and wi = 0 for i = S + 1, ..., Q.
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Rationale
Eff(ξCEA
S ) = Obj(ξ∗
S)
Obj(ξCEA
S )
≥ Obj(ξ∗
)
Obj(ξCEA
S )
where Obj(·) is the objective criterion function.
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Simulation study
3 × 3 × 2/2/12 3 × 3 × 2/2/12
Locally Bayesian Locally Bayesian
D-optimal 99.65% 98.71% 99.80% 98.82%
A-optimal 99.04% 96.83% 99.11% 97.15%
V -optimal 98.69% 96.75% 97.52% 96.85%
Table: Lower bound of the CEA efficiency under different design setups
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Simulation study (Cont.)
3 × 3 × 2 × 2 × 2/3/15 4 × 3 × 3 × 2/3/12
Locally Bayesian Locally Bayesian
D-optimal 96.76% 98.67% 96.90% 97.78%
A-optimal 92.84% 97.20% 95.11% 96.03%
V -optimal 94.21% 97.30% 94.12% 96.77%
Table: Lower bound of the CEA efficiency under different design setups
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm
Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you
Tian Tian MSCS, UIC
A discussion on the performance of the Coordinate Exchange Algorithm

More Related Content

What's hot

Analysis of Educational Robotics activities using a machine learning approach
Analysis of Educational Robotics activities using a machine learning approachAnalysis of Educational Robotics activities using a machine learning approach
Analysis of Educational Robotics activities using a machine learning approachLorenzo Cesaretti
 
IPL: An Integration Property Language for Multi-Model Cyber-Physical Systems
IPL: An Integration Property Language for Multi-Model Cyber-Physical SystemsIPL: An Integration Property Language for Multi-Model Cyber-Physical Systems
IPL: An Integration Property Language for Multi-Model Cyber-Physical SystemsIvan Ruchkin
 
Optimization problems and algorithms
Optimization problems and  algorithmsOptimization problems and  algorithms
Optimization problems and algorithmsAboul Ella Hassanien
 
A0311010106
A0311010106A0311010106
A0311010106theijes
 
Synthesis of analytical methods data driven decision-making
Synthesis of analytical methods data driven decision-makingSynthesis of analytical methods data driven decision-making
Synthesis of analytical methods data driven decision-makingAdam Doyle
 
Cone crusher model identification using
Cone crusher model identification usingCone crusher model identification using
Cone crusher model identification usingijctcm
 
A Literature Survey of Benchmark Functions For Global Optimization Problems
A Literature Survey of Benchmark Functions For Global Optimization ProblemsA Literature Survey of Benchmark Functions For Global Optimization Problems
A Literature Survey of Benchmark Functions For Global Optimization ProblemsXin-She Yang
 
Finite Element Method
Finite Element MethodFinite Element Method
Finite Element MethodBharat sharma
 
AIAA-Aviation-2015-Mehmani
AIAA-Aviation-2015-MehmaniAIAA-Aviation-2015-Mehmani
AIAA-Aviation-2015-MehmaniOptiModel
 
173448987 se3-4-it-2008
173448987 se3-4-it-2008173448987 se3-4-it-2008
173448987 se3-4-it-2008homeworkping9
 
Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...André Gonçalves
 
Forecasting time series for business and operations data: A tutorial
Forecasting time series for business and operations data: A tutorialForecasting time series for business and operations data: A tutorial
Forecasting time series for business and operations data: A tutorialColleen Farrelly
 
IRJET- Domestic Water Conservation by IoT (Smart Home)
IRJET- Domestic Water Conservation by IoT (Smart Home)IRJET- Domestic Water Conservation by IoT (Smart Home)
IRJET- Domestic Water Conservation by IoT (Smart Home)IRJET Journal
 
Modeling of Granular Mixing using Markov Chains and the Discrete Element Method
Modeling of Granular Mixing using Markov Chains and the Discrete Element MethodModeling of Granular Mixing using Markov Chains and the Discrete Element Method
Modeling of Granular Mixing using Markov Chains and the Discrete Element Methodjodoua
 
Verifications and Validations in Finite Element Analysis (FEA)
Verifications and Validations in Finite Element Analysis (FEA)Verifications and Validations in Finite Element Analysis (FEA)
Verifications and Validations in Finite Element Analysis (FEA)Kartik Srinivas
 
Types of device models(part 2)
Types of device models(part 2)Types of device models(part 2)
Types of device models(part 2)SudharshaG
 
Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...Seunghyun Hwang
 

What's hot (18)

Analysis of Educational Robotics activities using a machine learning approach
Analysis of Educational Robotics activities using a machine learning approachAnalysis of Educational Robotics activities using a machine learning approach
Analysis of Educational Robotics activities using a machine learning approach
 
IPL: An Integration Property Language for Multi-Model Cyber-Physical Systems
IPL: An Integration Property Language for Multi-Model Cyber-Physical SystemsIPL: An Integration Property Language for Multi-Model Cyber-Physical Systems
IPL: An Integration Property Language for Multi-Model Cyber-Physical Systems
 
Optimization problems and algorithms
Optimization problems and  algorithmsOptimization problems and  algorithms
Optimization problems and algorithms
 
A0311010106
A0311010106A0311010106
A0311010106
 
Synthesis of analytical methods data driven decision-making
Synthesis of analytical methods data driven decision-makingSynthesis of analytical methods data driven decision-making
Synthesis of analytical methods data driven decision-making
 
Cone crusher model identification using
Cone crusher model identification usingCone crusher model identification using
Cone crusher model identification using
 
A Literature Survey of Benchmark Functions For Global Optimization Problems
A Literature Survey of Benchmark Functions For Global Optimization ProblemsA Literature Survey of Benchmark Functions For Global Optimization Problems
A Literature Survey of Benchmark Functions For Global Optimization Problems
 
Finite Element Method
Finite Element MethodFinite Element Method
Finite Element Method
 
AIAA-Aviation-2015-Mehmani
AIAA-Aviation-2015-MehmaniAIAA-Aviation-2015-Mehmani
AIAA-Aviation-2015-Mehmani
 
173448987 se3-4-it-2008
173448987 se3-4-it-2008173448987 se3-4-it-2008
173448987 se3-4-it-2008
 
Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...Online learning in estimation of distribution algorithms for dynamic environm...
Online learning in estimation of distribution algorithms for dynamic environm...
 
AR model
AR modelAR model
AR model
 
Forecasting time series for business and operations data: A tutorial
Forecasting time series for business and operations data: A tutorialForecasting time series for business and operations data: A tutorial
Forecasting time series for business and operations data: A tutorial
 
IRJET- Domestic Water Conservation by IoT (Smart Home)
IRJET- Domestic Water Conservation by IoT (Smart Home)IRJET- Domestic Water Conservation by IoT (Smart Home)
IRJET- Domestic Water Conservation by IoT (Smart Home)
 
Modeling of Granular Mixing using Markov Chains and the Discrete Element Method
Modeling of Granular Mixing using Markov Chains and the Discrete Element MethodModeling of Granular Mixing using Markov Chains and the Discrete Element Method
Modeling of Granular Mixing using Markov Chains and the Discrete Element Method
 
Verifications and Validations in Finite Element Analysis (FEA)
Verifications and Validations in Finite Element Analysis (FEA)Verifications and Validations in Finite Element Analysis (FEA)
Verifications and Validations in Finite Element Analysis (FEA)
 
Types of device models(part 2)
Types of device models(part 2)Types of device models(part 2)
Types of device models(part 2)
 
Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...Your Classifier is Secretly an Energy based model and you should treat it lik...
Your Classifier is Secretly an Energy based model and you should treat it lik...
 

Similar to A discussion on the performance of the CEA (slides)

LNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine LearningLNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine Learningbutest
 
Learning to Search Henry Kautz
Learning to Search Henry KautzLearning to Search Henry Kautz
Learning to Search Henry Kautzbutest
 
Learning to Search Henry Kautz
Learning to Search Henry KautzLearning to Search Henry Kautz
Learning to Search Henry Kautzbutest
 
Sampling-Based Planning Algorithms for Multi-Objective Missions
Sampling-Based Planning Algorithms for Multi-Objective MissionsSampling-Based Planning Algorithms for Multi-Objective Missions
Sampling-Based Planning Algorithms for Multi-Objective MissionsMd Mahbubur Rahman
 
Numerical and experimental investigation on existing structures: two seminars
Numerical and experimental investigation on existing structures: two seminarsNumerical and experimental investigation on existing structures: two seminars
Numerical and experimental investigation on existing structures: two seminarsPhD ISG, Sapienza University of Rome
 
Accelerated life testing
Accelerated life testingAccelerated life testing
Accelerated life testingSteven Li
 
The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)theijes
 
Trajectory Transformer.pptx
Trajectory Transformer.pptxTrajectory Transformer.pptx
Trajectory Transformer.pptxSeungeon Baek
 
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...IJERA Editor
 
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATION
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATIONMODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATION
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATIONcscpconf
 
Modified Vortex Search Algorithm for Real Parameter Optimization
Modified Vortex Search Algorithm for Real Parameter Optimization Modified Vortex Search Algorithm for Real Parameter Optimization
Modified Vortex Search Algorithm for Real Parameter Optimization csandit
 
AMS_Aviation_2014_Ali
AMS_Aviation_2014_AliAMS_Aviation_2014_Ali
AMS_Aviation_2014_AliMDO_Lab
 
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...Cemal Ardil
 
Partha Sengupta_structural analysis.pptx
Partha Sengupta_structural analysis.pptxPartha Sengupta_structural analysis.pptx
Partha Sengupta_structural analysis.pptxJimmyPhoenix2
 
Slides TSALBP ACO 2008
Slides TSALBP ACO 2008Slides TSALBP ACO 2008
Slides TSALBP ACO 2008Manuel ChiSe
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchGreg Makowski
 

Similar to A discussion on the performance of the CEA (slides) (20)

LNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine LearningLNCS 5050 - Bilevel Optimization and Machine Learning
LNCS 5050 - Bilevel Optimization and Machine Learning
 
Learning to Search Henry Kautz
Learning to Search Henry KautzLearning to Search Henry Kautz
Learning to Search Henry Kautz
 
Learning to Search Henry Kautz
Learning to Search Henry KautzLearning to Search Henry Kautz
Learning to Search Henry Kautz
 
Sampling-Based Planning Algorithms for Multi-Objective Missions
Sampling-Based Planning Algorithms for Multi-Objective MissionsSampling-Based Planning Algorithms for Multi-Objective Missions
Sampling-Based Planning Algorithms for Multi-Objective Missions
 
Numerical and experimental investigation on existing structures: two seminars
Numerical and experimental investigation on existing structures: two seminarsNumerical and experimental investigation on existing structures: two seminars
Numerical and experimental investigation on existing structures: two seminars
 
Accelerated life testing
Accelerated life testingAccelerated life testing
Accelerated life testing
 
The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)The International Journal of Engineering and Science (The IJES)
The International Journal of Engineering and Science (The IJES)
 
Simulation Software Performances And Examples
Simulation Software Performances And ExamplesSimulation Software Performances And Examples
Simulation Software Performances And Examples
 
Trajectory Transformer.pptx
Trajectory Transformer.pptxTrajectory Transformer.pptx
Trajectory Transformer.pptx
 
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...
An Interactive Decomposition Algorithm for Two-Level Large Scale Linear Multi...
 
lecture.ppt
lecture.pptlecture.ppt
lecture.ppt
 
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATION
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATIONMODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATION
MODIFIED VORTEX SEARCH ALGORITHM FOR REAL PARAMETER OPTIMIZATION
 
Modified Vortex Search Algorithm for Real Parameter Optimization
Modified Vortex Search Algorithm for Real Parameter Optimization Modified Vortex Search Algorithm for Real Parameter Optimization
Modified Vortex Search Algorithm for Real Parameter Optimization
 
AMS_Aviation_2014_Ali
AMS_Aviation_2014_AliAMS_Aviation_2014_Ali
AMS_Aviation_2014_Ali
 
JOURNALnew
JOURNALnewJOURNALnew
JOURNALnew
 
P1121133727
P1121133727P1121133727
P1121133727
 
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...
Multi objective-optimization-with-fuzzy-based-ranking-for-tcsc-supplementary-...
 
Partha Sengupta_structural analysis.pptx
Partha Sengupta_structural analysis.pptxPartha Sengupta_structural analysis.pptx
Partha Sengupta_structural analysis.pptx
 
Slides TSALBP ACO 2008
Slides TSALBP ACO 2008Slides TSALBP ACO 2008
Slides TSALBP ACO 2008
 
Heuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient searchHeuristic design of experiments w meta gradient search
Heuristic design of experiments w meta gradient search
 

A discussion on the performance of the CEA (slides)

  • 1. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you A discussion on the performance of the Coordinate Exchange Algorithm Tian Tian Department of Mathematics, Statistics and Computer Science University of Illinois at Chicago 09 February 2016 Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 2. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Background of DCEs Background A product/service of interest K attributes each with lk levels L = K k=1 lk alternatives/profiles Each choice set consists of J profiles Q = L J choice sets Each design consists of S choice sets M = Q S candidate designs (design space) Goal –to estimate the appeal of each attribute –to predict the behaviors of consumers –to emulate market decisions and forecast market demands Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 3. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Background of DCEs Two critical components Parameter setup in the probabilistic model (we model the probability a consumer chooses profile j in choice set s — pjs) An efficient algorithm that finds the optimal discrete choice design(s) Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 4. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Challenges and motivation Challenges First challenge: The probabilistic choice models are nonlinear ⇒ efficiency of the design depends on the unknown parameters Possible solutions: Linear optimal designs where parameters are set to zeros Locally optimal designs where parameters are assigned with nonzero values Bayesian optimal designs where parameters are introduced with a prior distribution Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 5. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Challenges and motivation Challenges (Cont.) Second challenge: Difficult (impossible) to theoretically construct locally/Bayesian op- timal designs under a nonlinear multi-dimensional probabilistic choice model. Existing algorithms: The mFA (modified Federov algorithm) The RSA (relabeling and swapping algorithm) The RSCA (relabeling, swapping, and cycling algorithm) The quadrature scheme The CEA (coordinate-exchange algorithm) Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 6. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Challenges and motivation Motivation Question: Is the resulting design generated from the CEA indeed optimal, or at lease highly efficient? An example: 3 × 3 × 2/3/8 ⇒ (3×3×2 3 ) 8 > 4 × 1018 Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 7. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Review of the MNL model The MNL model The MNL (multinormial logit) probability pjs, j = 1, ..., J, s = 1, ..., S, is modeled as follows, pjs = exp(xjsβ) J i=1 exp(xisβ) – ˜K = K k=1(lk −1) is the total number of main effect levels subject to the ”zero-sum” constraint – xjs is a ˜K × 1 vector denoting the attribute levels of profile j in the sth choice set – β is a ˜K × 1 vector of parameter representing the main effects of the attribute levels Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 8. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Review of the CEA The CEA 1 Start with a randomly chosen starting design 2 Exchange levels for each attribute in each profile in each choice set in the design 3 A level is updated when the new design results in a smaller criterion value 4 A complete cycle of iteration terminates when the algorithm has reached the last attribute in the last profile in the last choice set 5 The algorithm goes back to the very first attribute and contin- ues another round Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 9. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Review of the CEA The CEA (Cont.) 6 A complete try of iteration terminates when no substitution is made in a whole cycle or other stopping rule is met 7 To avoid poorly local optima, T tries of the algorithm are re- peated each with a different starting design 8 The optimal design is chosen from the T resulting designs. Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 10. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Discrete and approximate design To take advantage of the structure under the continuous/approxi- mate design framework. Approximate/continuous: ˜ξ = {(Cq, wq)|q = 1, ..., Q}. Discrete/exact: ξ = {Cs|s = 1, ..., S} = {(Cq, wq)|q = 1, ..., Q} with wi = 1/S for i = 1, ..., S and wi = 0 for i = S + 1, ..., Q. Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 11. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Rationale Eff(ξCEA S ) = Obj(ξ∗ S) Obj(ξCEA S ) ≥ Obj(ξ∗ ) Obj(ξCEA S ) where Obj(·) is the objective criterion function. Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 12. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Simulation study 3 × 3 × 2/2/12 3 × 3 × 2/2/12 Locally Bayesian Locally Bayesian D-optimal 99.65% 98.71% 99.80% 98.82% A-optimal 99.04% 96.83% 99.11% 97.15% V -optimal 98.69% 96.75% 97.52% 96.85% Table: Lower bound of the CEA efficiency under different design setups Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 13. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Simulation study (Cont.) 3 × 3 × 2 × 2 × 2/3/15 4 × 3 × 3 × 2/3/12 Locally Bayesian Locally Bayesian D-optimal 96.76% 98.67% 96.90% 97.78% A-optimal 92.84% 97.20% 95.11% 96.03% V -optimal 94.21% 97.30% 94.12% 96.77% Table: Lower bound of the CEA efficiency under different design setups Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm
  • 14. Problem Introduction Review of the MNL model and the CEA Rationale of the approach Simulation study Thank you Tian Tian MSCS, UIC A discussion on the performance of the Coordinate Exchange Algorithm