PLS factoriel discriminant analysis (PLS-FDA) with cross-validation: a PLS2 (Simpls) computes scores from the disjonctive matrix of the classes, then a factoriel discriminant analysis is applied to the scores, yielding new scores; each observation is attributed to a class using these new scores
model = plsfda(x,y_class,split,lv,(metric),(scale),(threshold) )
a matrix (n x q) or a Div structure
a conjunctive vector (n x 1) or a disjonctive matrix (n x nclass) or a Div structure
for the cross-validation; the number of blocks or a vector of dimensions (n x 1) which identifies by a number the group of cross-validation of each observation
numbers of dimensions used for the PLS2 and the FDA
lv is a vector of dimensions 2; lv(1) is the dimension pf the PLS2, lv(2) is the dimension of the FDA; lv(2) should be lower or equal to nclass-1
the metric used for the calculation of the distances between the observations and the samples
metric=0: Mahalanobis distance (by default)
metric=1: usual Euclidian distance
the columns of x are always centered; scale defines the standardization
scale='c': no standardization
scale='cs': standardisation; the columns of x are divided by their standard deviation (by default)
the lowest value for attributing an observation to a class (by default= 1/nclass )
the confusion matrices for calibration, expressed as numbers of observations
model.conf_cal is a list of lv objects, each of dimensions (nclass x nclass)
the confusion matrices for calibration, expressed as percentages
model.conf_cal is a list of lv objects, each of dimensions (nclass x nclass)
the confusion matrices for cross-validation, expressed as percentages
model.conf_cv is a list of lv objects, each of dimensions (nclass x nclass)
the percentage of classification errors for calibrations (1st column) and cross-validations (2nd column)
model.err.d is a matrix (lv x 2)
the percentage of calibration error, for each class
model.errbycl_cal.d is a matrix (lv x nclass)
the percentage of cross-validation error, for each class
model.errbycl_cv.d is a matrix (lv x nclass)
the percentage of not-classed observations (all predictions lower than the threshold)
model.notclassed.d is a vector (lv x 1)
the percentage of not-classed observations (all predictions lower than the threshold), for each class
model.notclassed_bycl.d is a matrix (lv x nclass)
the discriminant method; here: 'plsda'
the original dataset (x); a div structure
model.xcal.d is of dimensions (n x q)
the original classes (y_class)
model.ycal.d is of dimensions (n x nclass)
the observations scores from Simpls; a div structure
model.scores.d is of dimensions (n x lv)
the R matrix of simpls, verifying: T=XR
model.rloadings.d is a matrix of dimensions (q x lv)
the metric used for the calculation of the distances between observations and groups
the applied standardization
the applied threshold