ModelOut structure

The ModelOut structure contains all the information relative to a model and the corresponding statistics when any.

The part in common for the different models is created by the function DefineModelOut. It contains the following fields:

Note In future versions of CuBatch this whole field may become obsolete as the both the ModelIn data structure and the X data structure may be kept within a Model file and the distinction between Models and Sessions technically abandoned.

This structure is model dependent and, apart from the field 'algorithm', is defined within the *Model functions or equivalents.
The 'algorithm' field contains the additional information associated to the computed model. Its content also are set by the *Model function responsible for handling the computation of the model.

Click on the corresponding link to check the content of the different 'info' structures

Contains the model's parameters as well as some statistics and quality parameters associated to it.
This field is a structure or a vector of structures (for some models such as PARAFAC and (n)PLS1).
In the latter case each element is associated to a different dimensionality of the model.

Fieldname	Content	Class
bcoeff	Regression coefficients	array of doubles
core	Core array	array of doubles
hotelling	Obsolete?	Obsolete?
nbfactors	number of factors/components/latent variables of the model	double
stats	statistics associated to the model.^§	structure
xcumpress	X CUMulated Prediction REsidual Sum of Squares (also known as RSS)	double
xev	X % explained variation	double
xfactors	loading matrices for the X array	cell vector of doubles.In most of the models the number of elements correspond to the number of dimensions of the X array
xpred	predictions for the X array	array of doubles of the same dimensions of the X used to compute the model.
xpreproc	X preprocessing parameters' structure with two fields: 'cen' and 'scal'. The first refers to centring and the second to scaling. Their contents are directly obtained via the nprocess function of the N-way toolbox.	The two fields are cell vectors where the n-th element refer to the preprocessing in the n-th mode. In case no preprocessing has been performed it is a cell vector of empty arrays.
xpress	X horizontal slabs (i.e. samples/batches) Prediction REsidual Sum of Squares	vector of double. The n-th element refers to the n-th slab
xrmse	X Root Mean Squared Error. It is computed over the existing (non-missing, that is) values only	double
ycumpress	Y CUMulated Prediction REsidual Sum of Squares (also known as RSS)	double
yev	Y % explained variation	double
yfactors	loading matrices for the Y array	cell vector of doubles. In most of the models the number of elements correspond to the number of dimensions of the Y array
ypred	predictions for the Y array	array of doubles of the same dimensions of the Y used to compute the model.
ypreproc	Y preprocessing parameters' structure with two fields: 'cen' and 'scal'. The first refers to centring and the second to scaling. Their contents are directly obtained via the nprocess function of the N-way toolbox.	The two fields are cell vectors where the n-th element refer to the preprocessing in the n-th mode. In case no preprocessing has been performed it is a cell vector of empty arrays.
ypress	Y horizontal slabs (i.e. samples/batches) Prediction REsidual Sum of Squares	vector of double. The n-th element refers to the n-th slab
yrmse	Y Root Mean Squared Error. It is computed over the existing (non-missing, that is) values only	double

^§ The fields of this structure are to be defined clearly and once and for all at the meeting taking place in Brussels on the 23 Oct. 2002

String with the model's name. The values admitted are the same as the Available_Models variable defined in AvModels

Vector of PlotStruct structures of the type.
Its final function has been fully implemented yet. Each of the elements should represent a possible plot available in the plant mode as defined by an advanced user.

Contains the predicted values for a new set of samples plus some additional information and statistics.
The fields of this structure and their contents are the following:

Fieldname	Content	Class
core	Core array Obsolete?	array of doubles
data	Cbdataset with the set of data upon which the model is applied	Cbdataset (double)?
nbfactors	number of factors variables of the model	double.
stats	Statistics associated to the predictions.^§	structure.
xcumpress	X CUMulated Prediction REsidual Sum of Squares	double.
xev	X % explained variation in prediction (Q²)	double
xfactors	Predicted X scores matrix	cell element containing an array of double.
xpred	Predictions for the X array	array of doubles of the same dimensions of the 'data' field.
xpress	X horizontal slabs (i.e. samples/batches) Prediction REsidual Sum of Squares	vector of double. Each element refers to one sample.
xrmse	X Root Mean Squared Error. It is computed over the existing (non-missing, that is) values only	double
yfactors	Predicted Y scores.	cell element containing an array of double.
ypred	Predictions for the Y array	array of doubles.

Contains the model's parameters as determined during the validation procedure.
It also contains some statistics and quality parameters depending on the validation.
This field is a structure or a vector of structures (for some models such as PARAFAC and (n)PLS1).
In the latter case each element is associated to a different dimensionality of the model.
The content of the fields may vary according to the employed method.

Fieldname	Content	Class	Ext. ^§§
bcoeff	Regression coefficients	array of doubles	Yes
core	Core array	array of doubles	Yes
method	Four types of validation are currently available (Table 1) Leave One Out Naïve Bootstrap Residuals Bootstrap Test Set validation	string of chars 'loo' 'nboo' 'rboo' 'test'	No
nbfactors	number of factors/components/latent variables of the model	double	No
segments	Its content depend on the methods. For 1.- 3. each column contains the indexes of the samples/batches employed for a certain sub-model. I.e. 'loo': the number of columns of this field is equal to the number of samples. In the n^th column contains the values 1 to n and n + 1 (if it exist) to I (i.e. the number of samples) '*boo': the number of columns is equal to the number of replicates. 'test': one column vector with the samples included in the test set (only).	array or vector of doubles	No
stats	statistics associated to the model.^§	structure	No
xcumpress	X CUMulated Prediction REsidual Sum of Squares (also known as RSS)	double	No
xev	X % explained variation	double	No
xfactors	loading matrices for the X array	cell vector of doubles. In most of the models the number of elements correspond to the number of dimensions of the X array	Yes^¤
xpred	predictions for the X array	array of doubles of the same dimensions of the X used to compute the model.	No
xpress	X horizontal slabs (i.e. samples/batches) Prediction REsidual Sum of Squares	vector of double. Each element refers to one segment.	No
xrmse	X Root Mean Squared Error. It is computed over the existing (non-missing, that is) values only	double	No
ycumpress	Y CUMulated Prediction REsidual Sum of Squares (also known as RSS)	double	No
yev	Y % explained variation	double	No
yfactors	loading matrices for the Y array	cell vector of doubles. In most of the models the number of elements correspond to the number of dimensions of the Y array	Yes^¤
ypred	predictions for the Y array	array of doubles of the same dimensions of the Y used to compute the model.	Yes^¤
ypress	Y horizontal slabs (i.e. samples/batches) Prediction REsidual Sum of Squares	vector of double. The n-th element refers to the n-th slab	No
yrmse	Y Root Mean Squared Error. It is computed over the existing (non-missing, that is) values only	double	No

^§ The fields of this structure are to be defined clearly and once and for all at the meeting taking place in Brussels on the 23 Oct. 2002

^§§When a resampling method is used (i.e. Leave One Out or Bootstrap) one mode is added to the double arrays of this field (when it is a cell this is valid for each elements).

^¤In 'loo' a further slab is added in the third dimension in the first mode (e.g the scores or the predictions for y) containing the predictions for the left out samples.