Release Notes Version 9 3

From Eigenvector Research Documentation Wiki
Jump to navigation Jump to search

Changes and Bug Fixes in Version 9.3

Version 9.3 of PLS_Toolbox and Solo was released in December, 2023.

(back to Release Notes PLS Toolbox and Solo)

New Features in Solo and PLS_Toolbox

  • ALS_SIT - Command line function for alternating least squares with shift invariant tri-linearity model
  • CLSTI - Add CLS Temperature Interpreted model type. Interpolates a test temperature from a give set of pure spectra
  • CROSSVAL - Added cross-validation by classes and stratified cross-validation
  • EVRISHAPLEY - Add Shapley values as additional variable importance measure and model explanation tool
  • LDA - Add Linear Discriminant Analysis
  • SPLITCALTEST - Added duplex, spxy, and random split methods

Other Features and Improvements

File Comment
  • Load .parent model of prediction automatically.
  • Make results reproducible by adding option, random_state.
DataSet Object
  • Update syntax to classdef. NOTE: this change will impact reverse compatibility.
  • Fix for for Windows pip hanging in Python configuration.
  • Add method to retrieve SSQ table from model object in multiple formats. RMSEC now availalbe for most models where available.
  • Fix for adding UMAP or TSNE models to modeloptimizer.
  • Add R2-Q2 plot type.
  • Add Single Variable Test.
  • Add ability to use preprocessing in interface and model.
  • Allow support for 1 component models.
  • Calculating class probability using exact expression instead of interpolating on lookup table values eliminates very small errors due to interpolation (PLSDA, ANNDA, and ANNDLDA).
  • When building a Gray CLS model (a CLS model with CLS Residuals as the clutter source for GLSW or EPO), allow plotting of the original CLS loadings and the clutter factor loadings. See this wiki page for more information on building a Gray CLS model: Building Gray CLS model
tecator demo dataset
  • The contents of the 'NIR of Meat Samples' ('tecator') demo dataset have changed. This now includes the three Y variables, 'moisture', 'fat', and 'protein', instead of just the 'fat' variable, and there are now no samples duplicated between the calibration and test datasets.