Rank-5 Tensor as Input for Regression, How?

Hi *,

I’m facing a problem of finding out how to treat my data and how to train on it for a regression problem.

The data is a list of cells (may refer them as pixels if you wish) with a value per cell. These cells form a 5 layers object in real life as each layer can be represented by a matrix. Each data object has a value for it thus my final goal is given these 5 layers, predict a single value.

I would love to hear how you think I should approach this?

My thoughts:

  1. Construct a Rank-5 Tensor and try to train a CNN on it. Is it possible “out of the box” in fastai? Tried but didn’t managed to actually call the fit function
    RuntimeError: The size of tensor a (5) must match the size of tensor b (3) at non-singleton dimension 1

  2. Construct a Rank-2 Tensor with every layer on the diagonal thus having an image to work with as a usual cnn. Couple issues here, first, the layers have some relations between them, meaning the center value of each one is related to the next/previous. Second, if I do choose to do so, make it a grey scale img and every pixel value maps to the real value?
    The attached image show such a cage.

Thank you