Release Notes Version 8 0

From Eigenvector Documentation Wiki
Jump to: navigation, search

Version 8.0 of PLS_Toolbox and Solo was released in June, 2015.

For general product information, see PLS_Toolbox Product Page. For information on Solo, see Solo Product Page.

(back to Release Notes PLS Toolbox and Solo)

New Features in Solo and PLS_Toolbox

Multi-Block, and Model and Data Fusion Tool

Multiblock Tool - Interface to view, manipulate, and join data. Can be used for data and model fusion, or multi-block modeling.

  • Join multiple blocks of variables measured on the same samples (alignment based on labels, axis scales, or size).
  • Automatically align and join time-based blocks of data (based on time axis scale).
  • Optionally build models on one or more blocks and join outputs from those blocks (model fusion).
  • Choose and apply block-specific preprocessing before joining.
  • Save multiblock model to use to join new data, including application of defined preprocessing and models.
  • After building model from joined data, Analysis automatically splits loadings into component block segments for ease of interpretation.

Analysis and Models

  • MLSCA - Multi-level simultaneous component analysis method added.
  • ASCA - Add model of residuals to help assess fit.
  • Shortcuts to Data Fusion methods Multiblock Tool and Hierarchical Model Builder
  • Re-designed Analysis and Preprocessing menus for ease-of-use and consistency.
  • ANN now supports custom cross-validation.
  • PLSDA variance captured plot now available.
  • Add "Reduced" Q and T^2 statistics for all factor-based models (normalized to confidence limit.)
  • Add quick-access to Genetic Algorithm variable selection from the iPLS, iPLSDA, and Stepwise Selection interfaces.
  • confusionmatrix - Report additional quantities for each class: count, classification error, precision and F1 score.
    • Standardize terminology: TP = count, TPR = proportion (rate) for confusion matrix quantities, and labels shown.
  • simca Better handling of full-rank PCA sub-models (where Q residuals are zero.)
  • Nearest neighbor score distance now normalized to maximum calibration value (standard practice for inlier tests.)
  • "Data" drill-down button in scores and loadings plots now automatically drills into preprocessed data and X_hat (fit, residuals) data if any plots of those types are already open.
  • Add Q2Y and R2Y to statistics calculated for models (for comparison to other software)

Plotting

  • Additional context-menu options for managing line width and symbol size.
  • Add quick access to class symbol sets in context menu.
  • Significantly faster selection display and "linking" between figures.
  • Connect Classes and View Classes buttons now have drop-down menus to display options.
  • Added Compress X-axis Gaps (click for example) toolbar button Compressgapsbutton.png to remove gaps caused by excluded variables or samples.
  • Improve handling of zoom status in newer versions of Matlab.
  • Better handling of font sizes on different screen sizes and platforms.
  • Fix shifting control position issues with newer versions of Matlab.

Importers

  • Automatic reconciliation of mixed axis scales when importing multiple files (using matchvars). Data will automatically include as much of the original data as possible.
  • omnicreadr New importer for OMNICix HDF5 image files.
  • hjyreadr Support for importing on 64-bit Windows systems and for new LabSpec file formats.
  • textreadr and xlsreadr Improved handling of multiple file import using graphically-selected parsing options. Options selected on first file/sheet are now used on ALL subsequent files/sheets.

Preprocessing

  • glog Generalized Log Transform added to preprocessing options.
  • pqnorm Probabilistic Quotient Normalization added to preprocessing options.
  • Group Scale
    • Added option to disable mean centering of block (scale only).
    • Added easier selection of class set to use when identifying blocks.
  • Added "Block Variance Scaling" as new method based on gscale
  • EEM Flitering (flucut)
    • Added better Raman and Rayleigh filtering using interpolation (now the default).
    • Added support for blank subtraction (choose one sample as a blank).
  • glsw Clarified how ELS/EMM and EPO options are related
  • Add support for handling missing data in both normaliz and mscorr (median only).

Other Interfaces

  • Model Optimizer - Better handling of numeric data in comparison table, additional statistics, and improved handling of include field.
    • Add better support for model groupings in PLSDA and SVMDA within model optimizer.
    • Add support for more LWR options
    • LWR models: Add "Survey" button to Analysis window to automatically survey over a range of "Local Points"
  • Better help integration with newer version of Matlab.
  • Hierarchical Model Builder - Add vertical scrolling.
  • TrendTool Add "maximum between" option for markers: returns the maximum value between two markers (better identifies peak value when the peak shape may change)

Model Objects

  • Build and change history now captured in history field of Model Object.
  • Add .scoredistance and .esterror as virtual properties for models. These properties can now be accessed directly from models in PLS_Toolbox or Solo_Predictor scripts.


New Command-line Features and Functions

Misc New Functions

  • eemoutlier - New function for automatically removing outliers in fluorescence PARAFAC models.
  • glog Generalized Log variable scaling.
  • kurtosis - Added kurtosis statistic function to distribution fitting toolbox.
  • mlsca Multi-level Simultaneous Component Analysis.
  • multiblock Create or apply a multiblock model for joining data.
  • pqnorm Probability Quotient Normalization for samples.
  • skewness - Added skewness statistic function to distribution fitting toolbox.

Command-Line Changes

  • als - Sort components by variance captured (if no constraints otherwise defining order).
  • comparemodels - Report the mean (class count weighted) of Classification Error, Precision, F1 Score for classification models.
  • confusionmatrix - Report additional quantities for each class: count, classification error, precision and F1 score..
    • Standardize terminology: TP = count, TPR = proportion (rate) for confusion matrix quantities, and labels shown.
  • cov_cv - Changed from SVDS to SVD to improve behavior with nearly-rank-deficient cases.
  • crossval - Better handling of cross-validation when using PLSDA.
    • Convert plsda regression method input to be 'sim' (to speed it up).
    • Recognize when user has passed single-column (either logical or class) and force it to be multi-column logical.
  • histaxes - Fix for when NaN's are present in data.
  • jmlimit - Better handle degenerate cases when multiple confidence levels are requested (return VECTOR of zeros instead of single zero).
  • matchvars - Add option to input a cell array of dataset objects which will be joined after reconciling variables to make the least changes in data.
  • mdcheck - Allow use of KNN as data replacement method (replace missing data with data from sample(s) which are closest).