in order to achieve high accuracy, I want to implement my own loss functions. For example, as I have read in the original nbeats paper, there is a custom loss function already in use (OWA).
However, in the nbeats documentation I am told to use either MAPE or MASE as loss metric. I do not like to rely on those metrics, for numerous reasons.
There are several documentations on how to define a custom loss function in keras. This does not seem very challenging.
Nevertheless, I do not know, how to make the loss function available to R via the reticulate interface.
Is there a way to create a yardstick-like framework, in order to create custom metrics?
Hello Matt,
in order to achieve high accuracy, I want to implement my own loss functions. For example, as I have read in the original nbeats paper, there is a custom loss function already in use (OWA). However, in the nbeats documentation I am told to use either MAPE or MASE as loss metric. I do not like to rely on those metrics, for numerous reasons. There are several documentations on how to define a custom loss function in keras. This does not seem very challenging. Nevertheless, I do not know, how to make the loss function available to R via the reticulate interface. Is there a way to create a yardstick-like framework, in order to create custom metrics?