One-Class SVM node Expert

On the Expert tab of the One-Class SVM node, you can choose from Simple mode or Expert mode. If you choose Simple, all parameters are set with the default values as shown below. If you select Expert, you can specify custom values for these parameters. For further detail about these options, see http://scikit-learn.org/stable/modules/generated/sklearn.svm.OneClassSVM.html#sklearn.svm.OneClassSVM.

Stopping criteria. Specify the tolerance for stopping criteria. Default is 1.0E-3 (0.001).

Regression precision (nu). Bound on the fraction of training errors and support vectors. Default is 0.1.

Kernel type. The kernel type to use in the algorithm. Options include RBF, Polynomial, Sigmoid, Linear, or Precomputed. Default is RBF.

Specify Gamma. Select this option to specify the Gamma. Otherwise, the auto gamma will be applied.

Gamma. The Gamma setting is only available for the RBF, Polynomial, and Sigmoid kernel types.

Coef0. Coef0 is only available for the Polynomial and Sigmoid kernel types.

Degree. Degree is only available for the Polynomial kernel type.

Use the shrinking heuristic. Select this option to use the shrinking heuristic. This option is deselected by default.

Specify the size of the kernel cache (in MB). Select this option to specify the size of the kernel cache. This option is deselected by default. When selected, the default value is 200 MB.

Hyper-Parameter Optimization (Based on Rbfopt). Select this option to enable Hyper-Parameter Optimization based on Rbfopt, which automatically discovers the optimal combination of parameters so that the model will achieve the expected or lower error rate on the samples. For details about Rbfopt, see http://rbfopt.readthedocs.io/en/latest/rbfopt_settings.html.

Target. The objective function value (error rate of the model on the samples) we want to reach (for example, the value of the unknown optimum). Set to an acceptable value such as 0.01.

Max Iterations. Maximum number of iterations for trying the model. Default is 1000.

Max Evaluations. Maximum number of function evaluations for trying the model, where the focus is accuracy over speed. Default is 300.

The One-Class SVM node requires the scikit-learn© Python library. The following table shows the relationship between the settings in the SPSS® Modeler SMOTE node dialog and the Python algorithm.

Table 1. Node properties mapped to Python library parameters
Parameter name	Script name (property name)	Python API parameter name
Stopping criteria	`stopping_criteria`	`tol`
Regression precision	`precision`	`nu`
Kernel type	`kernel`	`kernel`
Gamma	`gamma`	`gamma`
Coef0	`coef0`	`coef0`
Degree	`degree`	`degree`
Use the shrinking heuristic	`shrinking`	`shrinking`
Specify the size of the kernel cache (number input box)	`cache_size`	`cache_size`
Random seed	`random_seed`	`random_state`