Leave-one-out cross-validation for PLS regression or discriminant analysis
pls_cv = plscv(x,y,vl,'da')
input:
x (samples x descriptors) for cross-validation
y (samples x variables) for regression or
(samples x classes) for discriminant analysis. Classes numbers must be >0.
vl (1 x 1) number of latent variables to compute in cross-validation
'da' (char) to indicate PLS-discriminant analysis (in PLS regression it is no used)
output:
pls_cv struct with:
Ypcv (samples x variables x vl) predicted variables or
(samples x classes x vl) predicted classes for cross-validation
Tcv (samples x vl) x-scores for cross-validation samples
For PLS-R:
RMSEcv (variables x vl) Root Mean Square Error for cross-validation
R2cv (variables x vl) Correlation Coefficient for cross-validation
For PLS-DA:
Succv (1 x vl) Success (%) of classification for cross-validation
--------------------------------------------------------------------------
Model for Partial Least Squares regression or discriminant analysis
pls_model = pls(x,y,vl,'da')
input:
x (samples x descriptors) for calibration
y (samples x variables) for regression or
(samples x classes) for discriminant analysis. Classes numbers must be >0
vl (1 x 1) number of latent variables to model
'da' (char) to indicate PLS-discriminant analysis (in PLS regression it is no used)
output:
pls_model struct with:
Data (struct) X and Y input, and classes (for PLS-DA)
VLvar (2 x vl) cumulative variance (%) explained by model for X and Y.
Ypc (samples x variables) predicted variables or
(samples x classes) predicted classes for calibration samples
T (samples x vl) x-scores
P (descriptors x vl) x-loadings
W (descriptors x vl) x-weights
U (samples x vl) y-scores
Q (variables x vl) y-loadings
B (descriptors x variables) regression vectors
B0 (1 x 1) regression intercept for (mean(y,1))-(mean(x,1))
Lo (samples x 1) leverages for samples
SR (samples x variables) studentized residuals for samples
Lv (variables x 1) leverages for variables
For PLS-R:
RMSEc (1 x variavles) Root Mean Square Error for calibration
R2c (1 x variavles) Correlation Coefficient for calibration
RMSEc_Yrand (1 x variavles) RMSEc for Y-randomization test (mean of 10 shuffles)
R2c_Yrand (1 x variavles) R2c for Y-randomization test (mean of 10 shuffles)
For PLS-DA:
Succ (1 x 1) Success (%) of classification for calibration samples
Succ_Yrand (1 x 1) Success (%) of classification for Y-randomization test
--------------------------------------------------------------------------
Variables or classes prediction using PLS model
pls_pred = plspred(x,model,y)
input:
x (samples x descriptors) new samples for prediction
model (struct) with PLS calibration parameters
y (samples x variables) for regression or
(samples x classes) for discriminant analysis. Classes numbers must be >0. (optional for model test)
output:
pls_pred struct with:
Yp (samples x variables) predicted variables or
(samples x classes) predicted classes for new samples
Tp (samples x vl) x-scores for new samples
For PLS-R:
RMSEp (1 x variavles) Root Mean Square Error for prediction (only if 'y' is supplied)
R2p (1 x variavles) Correlation Coefficient for prediction (only if 'y' is supplied)
For PLS-DA:
Sucp (1 x 1) Success (%) of classification for prediction (only if 'y' is supplied) |
