2
$\begingroup$

Is there a way to incorporate multiple targets into one loss?

Currently, I work with the Sequential() API, I guess this won't be sufficient....

I work with area predictions as targets. Each sample of the Dataset has a finite area to allocate. my regression often overestimates the total sample area... I want to implement an area restriction into a loss. This would be the sum of all true target values for a sample. I want to separately penalize the overestimation of the area.

I know Keras does automatically average losses like this :

1.trained on batch -->

2.for all targets: loss (target) -->

3.mean loss over all targets

But i need something like this:

1.trained on batch -->

2.loss(sum all predicted areas, sum of true areas)& for all targets: loss (target)-->

3.mean loss overall targets and areasumloss

The current solution was a feature engineering approach :

total area per sample got added to the feature list but did not enhance the result. The sum of individual areas is still overestimated

Is there a way to do this with the Sequential() API or do i need to use the Functional API?

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.