How to store non-aligned data?

data
1

I try to deal with the chemical dataset, which has a lot of small molecules with different atoms. How to use torchdict to store those non-aligned data, for example, coordinates. What should I set the batch_size?

for example, some data looks like this:

ndims = 3
data = {
    'xyz': torch.concatenate([torch.rand(10, ndims), torch.rand(7, ndims)...]),
    'energy': torch.tensor([1, 2, ...]),
    'batch_ud': torch.tensor([0] * 10 + [1] * 7)
}

Thanks a lot!

2

Even if you stored them in a batch, how would you consume this batch in a model? I don’t think non-aligned data is supported anywhere in pytorch.