Trains an SVM with optimal parameters

status = model.trainAuto(samples, responses)
status = model.trainAuto(csvFilename, [])
[...] = model.trainAuto(..., 'OptionName', optionValue, ...)

Input

• samples See the train method.
• responses See the train method.
• csvFilename See the train method.

Output

• status Success flag.

Options

• Data See the train method.
• KFold Cross-validation parameter. The training set is divided into KFold subsets. One subset is used to train the model, the others form the test set. So, the SVM algorithm is executed KFold times. default 10
• Balanced If true and the problem is 2-class classification then the method creates more balanced cross-validation subsets that is proportions between classes in subsets are close to such proportion in the whole train dataset. default false

• CGrid, GammaGrid, NuGrid, PGrid, CoeffGrid, DegreeGrid Iteration grid for the corresponding SVM parameter. A structure that represents the logarithmic grid range of SVM parameters. It is used for optimizing model accuracy by varying model parameters, the accuracy estimate being computed by cross-validation. It accepts a struct having the fields below. It also accepts a 3-element vector in which each parameter is specified in the same order as the supported struct:

  • minVal Minimum value of the model parameter
  • maxVal Maximum value of the model parameter
  • logStep Logarithmic step for iterating the model parameter

  The grid determines the following iteration sequence of the model parameter values: minVal * logStep.^(0:n), where n is the maximal index satisfying minVal*logStep^n < maxVal. The grid is logarithmic, so logStep must always be greater then 1.

  Defaults are:

  • 'CGrid' : struct('minVal',0.1, 'maxVal',500, 'logStep',5 )
  • 'GammaGrid' : struct('minVal',1e-5, 'maxVal',0.6, 'logStep',15)
  • 'PGrid' : struct('minVal',0.01, 'maxVal',100, 'logStep',7 )
  • 'NuGrid' : struct('minVal',0.01, 'maxVal',0.2, 'logStep',3 )
  • 'CoeffGrid' : struct('minVal',0.1, 'maxVal',300, 'logStep',14)
  • 'DegreeGrid': struct('minVal',0.01, 'maxVal',4, 'logStep',7 )

The method trains the SVM model automatically by choosing the optimal parameters C, Gamma, P, Nu, Coef0, Degree of an SVM model. Parameters are considered optimal when the cross-validation estimate of the test set error is minimal.

If there is no need to optimize a parameter, the corresponding grid step should be set to any value less than or equal to 1. For example, to avoid optimization in gamma, set logStep = 0 in GammaGrid, and minVal, maxVal as arbitrary numbers. In this case, the value of the parameter Gamma is taken for gamma.

And, finally, if the optimization in a parameter is required but the corresponding grid is unknown, you may set it by name to obtain the default grid of that parameter. To generate a grid, for example, for gamma, set GammaGrid='Gamma'.

This function works for the classification (C_SVC or NU_SVC) as well as for the regression (EPS_SVR or NU_SVR). For ONE_CLASS, no optimization is made and the usual SVM with the specified parameters is executed.