Training a network detect int number from a set of numbers

I try to use pytorch to solve this issue:

  • training a network to detect int number from a set of numbers (my dataset with large int numbers for training)
  • use just CPU not CUDA feature (my video card has just CUDA 3.0 version and I can’t use it with pytorch)
    As I already said, I have a set of int numbers, but I need a good network.

If you have the inputs with their corresponding targets, I would just start with a simple model using two linear layers and a nonlinearity between them, e.g. F.relu.
I’m currently not sure, what kind of number should be detected, so could you explain your use case a bit more?

Something like a loto game (I don’t know all torchvision.models), but the network will be set by the good result versus bad.
The all Relu formula did not allow because is too simple and errors will follow a wrong way.
N-dimensional network will be more precise into training mode.