Custom training about ocaml-torch HOT 1 OPEN

There is no direct integration of csv reading in this package, however you can easily create tensors from arbitrary arrays of data so the easiest would probably be to use some other package to read the csv files, e.g. ocaml-csv, extract the relevant data and label and convert it to a tensor. If this process ends up being slow, the final tensors can be saved to disk via Tensor.save.
An alternative would be to read the csv files using python and produce .npz files via numpy. The upside being that some code is probably already available to read the csv files and extract the tensors there. Then the .npz files can be loaded directly from ocaml-torch (or converted to .ot files first via tensor-tools).

from ocaml-torch.