First of all, forgive me if I don't articulate this problem very well, I don't have a background in math.
I'm trying to model with machine learning an accumulation of values, starting at zero, that go to +/- infinity over enough time. These values are increased/decreased by some arbitrary (though meaningful) real number.
Right now, I store these values in various arrays that represent certain aspects of an abstraction as learning occurs:
However, they take up a lot of memory, so I would like to replace them with a single model, such as a neural network. But, modeling an accumulation such as this isn't easy, nor is accurately translating the normalized output created by most models into their respective "actual" values.
I was wondering if any of you gifted maths folks could help with this problem?