Hello everybody,
assuming I trained a deep learning network for a regression problem. I divided my dataset into training, test and validation parts and achieved very good results.
My entire data set consists of the hatched areas from the attached image.
If I now query these areas with my model, I get very good results.
But how can I tell another person or a reinforcement learning agent that they can only query certain x,y combinations? The Reinforcement Learning Agent would like to query the entire area. This could lead him to select controls that are not in the valid model range.
I would like to have an additional information about the validity of the model next to the model output. How do you deal with this topic?