Is the loss is the same as the error in deep learning?
I feel it's the same but I'm maybe wong...
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
2
Usually loss and error are different concepts, but sometimes people conflate the two because conceptually, they're similar.
Loss functions measure the misfit of the model -- how much the model is wrong.
Error usually is shorthand for "error rate," the proportion of samples misclassified.
These two concepts are not necessarily the same. For example, cross-entropy loss can be any non-negative number, but the error rate is some number between 0 and 1.
Moreover, the error rate is not a differentiable function so it is not suitable for use in the back-propagation algorithm. But cross-entropy loss is differentiable, and is perfectly reasonable to use in back-prop.
-
$\begingroup$ To see if I under-fit or over-fit my data should I look at the total loss or about the error rate? error rate = nbSamplesMisclassified/ nbSample right? $\endgroup$– FractaleCommented Nov 14, 2018 at 5:05
-
1$\begingroup$ General questions about neural networks and overfitting have a number of threads on this website. Here's one that seems like a good place to start. stats.stackexchange.com/questions/131233/… You can find more using the search feature. $\endgroup$– Sycorax ♦Commented Nov 14, 2018 at 8:18