6- IV-Tucker
Tucker model is an N-way model. For a scientific overview of the method, click
here.
Some specifications:
- It does not allow missing data.
- "Num.factors X" is the place
where the user puts the N parameters of X parameters.
- "Num.factors Y" is the place
where the user puts the N parameters of Y parameters.
The user has to write "-1"
when he does not want to reduce the correponding mode (Tucker2/1 ).
- "Num. common modes"
is the number of modes where You want to do the projections. It is automatically
done on the first modes.
- "OK" button
executes Tucker model.A default plot is returned after run (unless an information
appears). The other plots are available in the menu "results".
- "bootstrap" button
excutes Tucker model on original and resampled (by bootstrap) data. The user
will obtain the same plot as preceeding, but with a bootstrap estimation of
stability of the model (convex hulls, etc...). Available only for 3D!
- "Preprocessing" button
opens a new window, see below.
- "Postprocessing"
button opens a new window, see below.
- "close" button
closes the IV-Tucker window.
- Validation window:(Available
only for 3D!).Cross-validation is effectued, cutting off alternetively the
slabs corresponding to each modality of the 1st mode. Prediction is also tested.After
having chosen the maximal parameters, the user has to click on the "OK"
button. A surface is plotted .
-------------------------------------------------------------------------
Submenus and sub-windows:
preprocess menu:
see parafac window. Here preprocessing
is ava ilable for X and Y.
post-processing window:
Tucker solutions are subspaces:we can
choose the basis of these subspaces , in order to have the most interpretable
solution. Basis can be found in order to maximise different criteria.
- maximise the diagonale of the core
(when the core is cubic)
- maximise the slice-wise diagonale of
the core (when the 2 first dimensions of the core are equal.
- maximise thesquared variance of the
core.
- No rotation:default, i.e. Singular
Value Decomposition of the columns of the loadings.
Bootstrap options:
- Bootstrap model : naive (resampling
the horizontal slabs of X) and residual (resampling the horizontal slabs of
the residuals of Tucker(X)).
- Bootstrap replicates: Number of replicates.