Models developed in The Unscrambler® X can be saved and used in real-time in conjunction with The Unscrambler® Process Pulse and with the Unscrambler® Prediction Engine and Classification Engine (which can be embedded into third party software applications). Models can be saved as the full model or in a smaller format, which allows you to make predictions and also report some of the other diagnostic calculations with predictions. To save a model as an independent file, right click on it and select Save Model which opens the following dialog. From here different model sizes can be chosen.
The prediction result options allow you to save the model as a smaller file which does not include all the matrices generated with the entire PLS model.
The prediction model options are:
Full with support for inlier detection: This model can be used for prediction, giving all the results that The Unscrambler® computes on prediction, including the deviation.
Full: This model results file allows you to predict new values, and get the deviation with that value, as well as to detect outliers (based on Hotelling’s T2 and Q residuals). With this model, inliers cannot be computed during the prediction stage.
Short: In the short model, only the raw beta coefficients are saved, at the optimal (or user-defined) number of components. No validation matrices are saved. With a short prediction model, you can get the predicted results for new data, but no other distance measure, or deviation measure. A short prediction model is not recommended if you would like to have model and/or sample diagnostics during the prediction step.