Add Parallelism

[fpicetti]

As the network size is proportional to the dataset size, and the GPU memory is limited, some parallelism has to be implemented.

A naive approach is to split the dataset into possibly-overlapped windows to be processed separately and then the results are merged. This can be done sequentially (as in this repo) or with a DataParallel strategy.

On the other hand, in order to process all the data at once, one might want to adopt a ModelParallel strategy.

[frailty-cyber]

data里面为什么没有hyperbolic3d_irregular_066.npy？

[fpicetti]

Hi @frailty-cyber,

Thanks for your comment! Indeed, the README needs to be updated, sorry for the delay.
You can find that file in datasets/hyperbolic3d/. We provide you with the original data volume, and two different decimations.

Let me know if you need any help. I would ask you to open a new issue, as this thread is dedicated to the parallelism.