Bayesian Optimization <undefined> and NaN Results

Question

MByk 2018-6-26

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/407503-bayesian-optimization-undefined-and-nan-results

编辑： MByk 2018-6-27

Sorry, if it's a silly question. I am using Bayesian Optimization to optimize classifier hyperparameters but sometimes I having "<undefined>" and "NaN" values for some parameters. What do they mean? Dataset is not suitable for this classifier? Should I use classifier's default parameters? Thanks for the help.

0 个评论
显示 -2更早的评论隐藏 -2更早的评论

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

Don Mathis 2018-6-26

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/407503-bayesian-optimization-undefined-and-nan-results#answer_326371

编辑：Don Mathis 2018-6-26

I would need to see your example to be sure, but a typical case is when some parameter is not used when some other parameter has a certain value. For example, the PolynomialOrder parameter of an SVM is only used when the KernelFunction parameter is 'polynomial'. So a NaN or "<undefined>" value in a parameter vector means that you should not use that parameter.

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

MByk 2018-6-26

编辑：MByk 2018-6-27

在 MATLAB Online 中打开

Wine.txt

Thank you very much.

X = DataSet(:,(1:end-1));
Y = DataSet(:,end);
Disp_Opts = struct('Optimizer','bayesopt','ShowPlots',false,...
                  'Verbose',1,'AcquisitionFunctionName','expected-improvement-plus');
Mdl_Eva = fitcnb(X,Y,'OptimizeHyperparameters','all',...
                    'HyperparameterOptimizationOptions',Disp_Opts);