Successive Projections Algorithm – Graphical User Interface
For use with Matlab 6.5 (R13) with the Statistics Toobox (function finv.m)
Type gui_spa at the Matlab prompt to start the graphical user interface.
“Main” screen
Select data file (.mat)
Press the “…” button, choose the .mat file containing all X and y matrices (calibration, validation,
and prediction) and press the “Load file” button
Line plot: select a matrix
Select one of the matrices that were loaded in the previous step, choose one or more rows (or
columns) and press the “Plot” button. To choose multiple rows (or columns), keep the Ctrl key
pressed.
“Successive Projections Algorithm” screen
If the mouse pointer is placed on top of a data input field, a brief help textbox will be displayed.
Press the “…” buttons and choose the X and y matrices for calibration and validation. If the
validation fields are left empty, leave-one-out cross-validation will be carried out.
The fields of minimum and maximum number of variables are optional. If left blank, default values
will be assumed.
The data are mean-centered prior to the projection procedures. If the “Autoscaling” checkbox is
selected, autoscaling will also be carried out.
Press “Run SPA” to run the algorithm. The indexes of the selected variables will be placed in the
workspace (array “var_sel”).
“Prediction Metrics” screen
Press the “…” buttons and choose the X and y matrices for calibration and prediction, as well as the
array with the indexes of the selected variables (which was given the name “var_sel” in the previous
screen).
Press the “Run Validation Metrics” button to obtain a Predicted vs Reference plot with a bisectrix
line and one-sigma confidence intervals, as well as the following statistics associated to prediction
performance:
PRESS (Prediction Errors Sum of Squares)
RMSEP (Root Mean Square Error of Prediction)
SDV (Standard Deviation of the Prediction Errors)
BIAS (Average of the Prediction Errors)
r (Correlation Coefficient between Predicted and Reference values)
If Xpred and ypred are left blank, leave-one-out cross-validation will be carried out.
