Writing custom loss in tensorflow that does not use Y_pred and Y_actual

My loss function for a DNN classification task is eigenvalue-based which does not need the inputs Y_prediction and Y_actual. Is it possible to write specialized custom loss functions like that using Tensorflow? >Solution : Of course: def customLoss(y_true, y_pred, alpha): loss = ….alpha return loss model.compile(loss=customLoss(alpha), optimizer=’sgd’)