Ytukyg

I'm trying to build a neural network with Keras. Therefore, I need to implement a custom loss function that punishes a false prediction example based on another attribute from the example.
I don't really know if it is even possible to use single attribute values inside a custom loss function in keras.
Here is the header of my data set.

Ereignisart;KONTOPHASE;PRODUKTART;PRODUKT2;BAUSPAREINLAGE;BAUSPARSUMME;BEWERTUNGSZAHL;TREUEOPTIONSSTATUS;ZUTEILUNGSSTATUS;MEHRZUTEILUNG_BETRAG;SPARZINSSATZ;Marktzins;ZEITSCHEIBE;SALDO

The first attribute "Ereignisart" is the one I want to predict. With the last attribute "SALDO" I want to punish the prediction of "Ereignisart" weighted. I know how to write a custom loss function and give it the attribute "SALDO" ( i got this from here if someone is interested: https://datascience.stackexchange.com/questions/28440/custom-conditional-loss-function-in-keras) . But if I do it the way you can see below, every example of SALDO will be used and not only the given value of the example.content of SALDO in debugger

def customLoss(SALDO):

def loss(yTrue, yPred):

    first_log=kBack.log(kBack.clip(yPred, kBack.epsilon(), None) + 1.)

    second_log=kBack.log(kBack.clip(yTrue, kBack.epsilon(), None) + 1.)

    return kBack.mean(kBack.square(first_log - "penalty dependent on SALDO" *second_log)), axis=-1)

return loss

In short, I guess I want to use the corresponding value of "SALDO" in my loss function for each pair of yTrue and yPred to punish a false prediction based on "SALDO".

tl;dr Is it possible to use single values of attributes in a custom Kears Loss function? If so, how?

Many thanks in advance

Fabian