fitcsvm cross-validation

Question

João Mendes 2021-4-15

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/803266-fitcsvm-cross-validation

评论： João Mendes 2021-4-16

Hi, I am training a SVM classifier with the following code:

SVM_1=fitcsvm(X_train, y_train, 'OptimizeHyperparameters', 'all','HyperparameterOptimizationOptions',struct('Optimizer','bayesopt','AcquisitionFunctionName','expected-improvement-per-second-plus','Kfold',10,'ShowPlots',0)); 

I was wondering if there is any possibility to retrieve a performance metric of the classifier from the cross-validation - since I specify it as a 10-fold cross-validation (AUC, for example).

Thank you,

J

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

Alan Weiss 2021-4-16

0
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/803266-fitcsvm-cross-validation#answer_676746

As shown in this doc example, the cross-validation loss is reported at the command line and plotted by default (I see that you turned off the plot). Is there something else that you need, or did I misunderstand you?

Alan Weiss

MATLAB mathematical toolbox documentation

3 个评论
显示 1更早的评论隐藏 1更早的评论

Alan Weiss 2021-4-16

The "Objective" in the iterative display (the generated table of iterations) is the cross-validation loss. The "Best so far" is simply the minimum objective up to that iteration. There is a difference between the "best so far" estimated and observed; that is a function of the model that the solver is estimating, and that changes every iteration. The model is that the observations themselves are noisy, so simply observing a value doesn't mean that observing it again will give the same response.

In a nutshell, I think that the iterative display gives you the information you seek.

Alan Weiss

MATLAB mathematical toolbox documentation

João Mendes 2021-4-16