0
$\begingroup$

When training a model (classifier) in TensorFlow, an accuracy value is returned. What is the interpretation of an accuracy of, say, 0.79. Furthermore, how does the accuracy relate to other evaluations of predictions, such as F1?

$\endgroup$
4
2
$\begingroup$

The accuracy is the proportion of correct predictions. To derive the F1 score, you need additional information drawn from the confusion matrix. The latter is easy to calculate with any classification model, see e.g., https://en.wikipedia.org/wiki/Confusion_matrix how to calculate a zillion statistics from it, including accuracy and F1 score.

$\endgroup$
4
  • $\begingroup$ Thanks! So also the F1 score (via the confusion matrix) is also based on the relationship between the training set and the validation set? We never calculate any accuracy score in terms of bringing in unseen data and evaluating that (right)? $\endgroup$
    – DIGSUM
    Nov 22 at 19:15
  • $\begingroup$ The F1 score is a function of the predicted and the observed classes of one single dataset, so I don't understand your comment. $\endgroup$
    – Michael M
    Nov 22 at 20:08
  • $\begingroup$ If we split a labelled dataset 20 (test) / 80 (train) and then get the F1 score from the confusion matrix, we get to know how well the model can predict, right? If we move on to let the model predict labels for unseen data, we don't have any way of knowing how well it performed on that unlabelled data, right? $\endgroup$
    – DIGSUM
    Nov 22 at 21:52
  • 1
    $\begingroup$ For unlabelled data, you can't calculate a confusion matrix, right. But you can derive one for your 80% training data and another one from your 20% "test" dataset. $\endgroup$
    – Michael M
    Nov 23 at 5:59

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.