Choosing an Alternative Meta Model for SPOT's Sequential Task

SPOT processes data sequentially, i.e., starting from a small initial design, further design points (parameter settings for the algorithm) are generated using a meta model. Many meta models are available in R. This is one of the main reasons why SPOT is implemented in R. The user can use state of the art meta models for tuning his algorithm. Or, he can write his own meta model and use it as a plugin for SPOT.

The default SPOT installation contains several meta models (and further meta models will be added in forthcoming versions). Table 6 summarizes meta models from the current SPOT version (0.1.888). The command spotVersion() displays the actual version of your local SPOT package.

randomForest was chosen as the default meta model, because it is quite robust and can handle categorical and numerical values.

Table 6: SPOT meta models

These plugins should be considered as templates. They were implemented in order to demonstrate how the interfaces should look like. We strongly recommend an adaptation of these plugins to your specific needs.