MATLAB Answers

0

Hyperparameter Optimization in ECOC classifier: which loss function is used?

Asked by Elena Casiraghi on 20 Sep 2019 at 11:45
Latest activity Commented on by Elena Casiraghi on 20 Sep 2019 at 19:15
Dear, I'm training an ECOC classifier using knn as the base classifier.
I would like to use the option 'OptimizeHyperparameters','auto' to let fitcecoc apply leave one out cross validation the best Coding, NumNeighbors, distace parameters.
tknn = templateKNN();
mdlknnCecoc = compact(fitcecoc(XKnn,labelsRed, ...
'OptimizeHyperparameters','all', ...
'HyperparameterOptimizationOptions',struct( 'UseParallel',...
true,'CVPartition',c), 'Learners',tknn));
In MATLAB help I read: " The optimization attempts to minimize the cross-validation loss (error) for fitcecoc by varying the parameters."
However, which loss function is used? I found no detail about that.

  0 Comments

Sign in to comment.

1 Answer

Answer by Don Mathis on 20 Sep 2019 at 13:55
 Accepted Answer

it says
"The optimization attempts to minimize the cross-validation loss (error) for fitcecoc by varying the parameters. For information about cross-validation loss in a different context, see Classification Loss. "
If you click on "Classification Loss" it tells you about the multiclass loss function.

  3 Comments

I did look at that help page. It describes all the loss functions that can be used. However, it doesn't say which loss is used
Yes, I see that now. The answer is 'classiferror', because that's the default loss for kfoldLoss for classification models.
When optimization is used, kfoldLoss is called with its default loss to compute the cross-validated loss to be optimized. The linked-to page was actually the classification kfoldLoss page, and if you scroll up you can find where it lists its default loss. I'm sorry it's not easier to find than that.

Sign in to comment.