Amazon Rekognition Model metrics are available for only test dataset, not train dataset. I think having access to training results ,not only testing results, can be of immense use to everyone using Rekognition. Even if we do not know or are unable to change model architecture(AWS internal team is responsible for this), one very simple but useful information to be gained from comparing train and test results could be( if the train F1 score > test score by a lot) ascertaining whether train and test distributions are vastly different or not. Of course many other things can also be inferred from this information. Hence, I sincerely believe gaining access to this sort of train metrics information will be useful to everyone and enhance user experience. Can anyone here associated with AWS internal team/working inform me why this is not a feature.
AWS OFICIALAtualizada há um ano