Exporting Models

From Eigenvector Documentation Wiki
Jump to: navigation, search

In most versions of PLS_Toolbox and Solo, there are four options for exporting models:

To File

Exporting to file saves the content of the model (all the internal numerical values in the model). This is similar to saving the model to a .mat file but it is generally "plain text". This is a good method to save models in a human-readable format but still requires PLS_Toolbox, Solo, or Solo_Predictor to apply the model.

Formats include .xml (text), .amo (ASCII-MOD), .m (text), or .mat (binary) formats. Some custom versions of Solo allow for other export options.

To Regression Vector

Exporting to a regression vector (only available for regression models) saves the model in the simplest predictor format. It outputs a mean vector and a regression vector which can be used with new data to make predictions. This does not include the numerical vectors needed to calculate T2, Q, scores, or contributions. You can only get the y-prediction with this option. This option only supports the basic mean centering and autoscaling preprocessing options.

When this option is selected, you will be prompted to save the regression vector and mean vector to one of three formats:

  • .mat - a binary format readable by Matlab.
  • .csv - a plain-text comma-separated-values format readable by Microsoft Excel as well as other programs.
  • .xml - a text format with XML tags enclosing each of the vectors.

For technical details on this option, see the PLS_Toolbox function Regcon.

To Predictor

Exporting to a predictor (which requires the Model_Exporter add-on product) produces a predictor script. This option is similar to exporting to a regression vector but the exported model:

  1. supports all preprocessing options in Analysis
  2. works on PCA, PLS, PCR models
  3. outputs scores, Q, T2, Q contributions, T2 contributions, AND predictions

You can get more information on Model_Exporter from here: Model Exporter User Guide

This format contains the code and numerical values necessary to make a prediction, but you must still have additional software to interpret that model. It does not support variable alignment, missing data, or other model types. For these options, see Solo_Predictor which supports all model types, calibration transfer, variable alignment, missing data support, and all preprocessing methods.

To Model Builder

This option outputs a [Matlab] m-file function which can be used with PLS_Toolbox and Matlab to build new models with the same conditions as the currently loaded model. This option takes all the Analysis GUI options and preprocessing settings necessary to reproduce the current model at the Matlab command line. It can be used to create new models from new data using the same conditions as the current model. For more technical details on this option, see the PLS_Toolbox function Encodemodelbuilder.

To OpenFluor.org

You can export a PARAFAC model of fluorescence spectra for querying the openfluor database http://www.openfluor.org. This option takes the dataset, the PARAFAC loadings, axisscales, and formats it into a .txt file for querying the database. When queried, the database returns the top matches. For more technical details on this option, see the PLS_Toolbox functions: queryopenfluor and parafacforopenfluor.

To Vision Air

This options exports an EVRI Model Structure to Vision Air PLT model. This option takes the EVRI model in the workspace and a name of the file to create and saves the model it as a PLT file. See the PLS_Toolbox function: writeplt.