> don't necessarily generalize to the whole population In my experience it is far more likely to mean "doesn't generalize well, and moving to a reasonable tra

> don't necessarily generalize to the whole population In my experience it is far more likely to mean "doesn't generalize well, and moving to a reasonable training set will reveal flaws in the model.