Index     


Models

You can choose between a number of different linear and non-linear regression models and some classification approaches.

Currently there are the following linear regression models:

  • Muliple Linear Regression (MLR)
  • Ridge Regression (RR)
  • Principle Component Regression (PCR)
  • Partial Least Squares (PLS)
  • Orthogonal Partial Least Squares (OPLS)

Weighted linear regressions:

  • Automated Lazy Learning (ALL)
  • K-Nearest Neighor Regression (KNN)

Non-linear, kernel-based, regression models:

  • Gaussian Process (GP)
  • Kernel Principle Component Regression (KPCR)
  • Kernel Partial Least Squares (KPLS)
  • Support Vector Regression (SVR)

Classification models:

  • Linear Discriminant Analysis (LDA)
  • Simple Naive Bayes (snB)
  • Naive Bayes (nB)

Creating a model

Linear regression models

In order to create a new model, just drag its item from the model panel onto an existing input item within the pipeline area (a).
You will then be shown a dialog that helps you set the necessary parameters.

The model's parameter can be set in dialog-page (b), whereas (c) allows to enable automatic optimization of model parameters by use of cross validation (where available).

Kernel-based regression models

In case of kernel-based non-linear regression models, you also need to choose the desired type of kernel function (a).

Currently you can select one of the following:

  • Polynomial kernel function
  • Radial basis kernel function
  • Sigmoid kernel function

A value for the parameter(s) of the chosen kernel function can be set in (b). These values can be optimized by grid search, so that the value entered here will serve only as a start-value if this is done.
For clearness the selected kernel-function will be displayed below (c).

If optimization of model and kernel-parameters is desired, it can be activated in dialog-page shown on the right.
The number of cross validation folds to be used for model and kernel-parameter optimization can be set in (d), while (e) lists the names of the model-parameters that will be optimized.
The number of grid search steps (f), the step-width (g) and the number of recursions for kernel-parameter optimization can be altered below. If the latter one if larger than 0, grid search is repeated as often as stated here. In each iteration the step-width is automatically reduced by a factor of two and the search is done locally around the best parameter(s) found in the previous iteration.


    Index