SPOT Configuration

A configuration (CONF) file, which stores information about SPOT specific settings, has to be set up. For example, the number of SANN algorithm runs, i.e., the available budget, can be specified via auto.loop.nevals. SPOT implements a sequential approach, i.e., the available budget is not used in one step. Evaluations of the algorithm on a subset of this budget, the so-called initial design, is used to generate a coarse grained meta model $F$. This meta model is used to determine promising algorithm design points which will be evaluated next. Results from these additional SANN runs are used to refine the meta model $F$. The size of the initial design can be specified via init.design.size. To generate the meta model, we use random forest (Breiman, 2001). This can be specified via seq.predictionModel.func = "spotPredictRandomForest".

Random forest was chosen, because it is a robust method which can handle categorical and numerical variables.