In a lot of machine learning blogs or review, the training dataset accuracy (or other metric) is given alongside the test dataset score. Is this score calculated through the training, or is the training data given at the final model to see if it gives a good score ?

Moreover, what is the point of computing such a score, as the model has been trained with this data ?



Computing the training score and the test score is usefull to detect overfitting

for exemple here we can see that the training score (in blue) keep improving but at some point the test score (in red) stop improving, that's the sign of overfitting.

from wikipedia

But for a final model the training score is not usefull and only the test score should be used.

|improve this answer|||||
  • $\begingroup$ Thanks, but is the training score computed alongside the training or is it computed at each epoch i.e. all the training set is ut through the current model at time t ? $\endgroup$ – Alexis Pister May 29 '19 at 10:28
  • $\begingroup$ for each epoch I compute the total loss of the epoch on the training set and the loss on the test set $\endgroup$ – vico May 29 '19 at 10:36

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.