"NumBins" error when using fitcsvm, setting at least one hyperparameter to optimize, and specifying Nu

Question

Dammak 2019-11-14

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/491119-numbins-error-when-using-fitcsvm-setting-at-least-one-hyperparameter-to-optimize-and-specifying

评论： Dammak 2019-11-28

采纳的回答： Athul Prakash

在 MATLAB Online 中打开

If I use fitcsvm with setting Nu and specifying at least one parameter to optimize, I get a non-informative error.

Here's sample code that will trigger the error:

features = [10;10;3;2];
labels = [1;1;0;0];
optHyp = hyperparameters('fitcsvm', features, labels)
optHyp(1).Optimize = true;
fitcsvm(features, labels ,'Nu',0.3, 'OptimizeHyperparameters',optHyp);

Here's the error I get:

This happens no matter what Nu value I set between 1 and 0, with different sets of features and labels, and with different hyperparameters getting set to optimize.

Furthermore, there is no user input called NumBins or similar for the classifier or the hyperparameters.

Also, the documentation doesn't show any dependence relationship between Nu and other parameters.

Anyone know what's going on here?