Accuracy Measure in Keras Network Learner Node

Hi, I have 2 questions for the Deep Learning experts on the forum:

**Which accuracy measure is used to display the accuracy in the Keras network learner monitor? **
… and can it be changed? (ie. from binary accuracy to top k accuracy)

I am currently working on a sparse multiclass multilabel classification problem.
The network layer structure is quite simple (input layer, dense layer -relu, dropout layer, dense layer -sigmoid), and the selected loss function is binary cross-entropy.
While the training monitor shows optimistically good results, scores from the validation set are more sobering (in line with my expectations). I assume the optimistic results in the learning monitor are due to the accuracy measure being set to binary accuracy…

Thanks for any insights on this,

M. Wakileh

Hi @mwakileh,
indeed node description lacks information on what accuracy measure is being used. To me it also looks like binary accuracy is being used, based on some multiclass experiments.
As of now there is no option to change the accuracy measure.
I will file tickets for both, adding information to node description and an option to adjust accuracy measure.