Multiway Data Analysis
  Tentative Programme
May 26-30, 2008  
 
 

Tentative Programme

Monday May 26

10:00h to 14:00h

Level: Application oriented / Non-technical level.

Exploratory Analysis of Two-way data (Principal component analysis, PCA).

Description of various types of three-way data (three-mode vs. two-mode).

Introduction into most popular methods for exploratory analysis of three-way data (CANDECOMP/PARAFAC, Tucker3), which are generalizations of PCA.


Tuesday May 27

10:00h to 14:00h

Level: Application oriented / Low-technical level.

Preprocessing three-way data (centering, normalizing).

Ins and outs of CANDECOMP/PARAFAC.

  • main properties of method
  • interpretation of solution
  • problematic cases
  • fitting procedures, fit measures
  • choice of dimensionality
  • stability checking

Level: Computer practice.

Introduction into matlab routines and Kiers´ three-way program.


Wednesday May 28

10:00h to 14:00h

Level: Application oriented / Low-technical level.

Ins and outs of Tucker models.

  • main properties of method
  • interpretation of solution
  • fitting procedures, fit measures
  • choice of numbers of components
  • stability checking
  • bootstrap analysis
  • plotting procedures
  • simplicity transformations
  • dealing with data with one extremely large mode

Level: Computer / Data analysis practice.

Carrying out example three-way analyses, using matlab routines.

Possibly: Analyzing data from participants own research.


Thursday May 29

10:00h to 14:00h

Level: Application oriented / Low-technical level.

Description of methods for analyzing two-mode three-way data (i.e. three-way data in which data slices refer to the same row and column entities, e.g. sets of correlation matrices among a number of variables, sets of matrices of similarities among a number of entities, etc.).

  • INDSCAL
  • IDIOSCAL
  • PARAFAC2
  • French three-way methods (AFM, Statis)

Level: Application oriented / Low-technical level.

Two kinds of special three-way methods:Constrained three-way methods.

  • General idea of constraining some parts of the solutions (e.g., requiring components to be equal, requiring certain values to be zero, etc)
  • Three-way clustering methods (i.e., methods in which the solution for one or more of the modes is given in terms of clusters rather than components, e.g. clusters of objects, clusters of variables, etc.)
  • Time constrained three-way methods (i.e., for data in which one of the modes refers to time points, this mode is described by smooth and/or time ordered components) Multiway regression analysisIn practice one frequently encounters the situation in which one set of data is used to predict or explain the values in a different set of data. Methods for modeling such relations between multiway data sets are multiway regression analysis methods.
  • Multiway regression analysis. In practice one frequently encounters the situation in which one set of data is used to predict or explain the values in a different set of data. Methods for modeling such relations are multiway regression analysis methods.

Friday May 30      Extension of Wednesday (by Jos ten Berge)

10:00h to 14:00h

Level: Most technical level so far. Targeted at chemometricians.

On Tucker-3 fitting with may core elements constrained to be zero. Mathematical results (based on typical rank and simplicity transformations) and Simplimax.

Level: Computer practice and presentation in the group.

Analyzing further examples from participants own research.