what is training error? what is testing error? what are their interpretations in the context of the data and what do they mean (in the context of the data)?