Normally when evaluating a model, the training data is split into a test set and a training set.

I want to evaluate the best possible performance of the classifier on my task.

So I have Trained it on all my data, and tested it on all my data. The results are not all 100% for all variation on the model I am investigating, which is worth reporting on.

  • $\begingroup$ You may have to provide more detail on how you developed your model (or whether it was a pre-existing model, for which you estimated the parameters on the training set) and what you did to avoid overfitting. Some approaches have well recognized names (e.g. leave-one-out cross-validation, x-fold cross-validation etc.). $\endgroup$ – Björn Aug 26 '15 at 12:04
  • $\begingroup$ I am also doing various forms of cross-validation. But not here. Here I want the classifier to overfit. The purpose of setting the test and training set to be the same is to assess if the classifier has the power to apply to this data at all. $\endgroup$ – Lyndon White Aug 26 '15 at 12:08
  • $\begingroup$ If your main approach to avoid overfitting is cross-validation, then I guess people would know what you did, if you wrote e.g. "naively applying the classifier without cross-validation". $\endgroup$ – Björn Aug 26 '15 at 14:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.