How to unbatch a big disconnected graph?

[BraveDistribution]

Hello,

I am working with meshes and I have a DataBatch, with batch size of 4. I am interested in splitting the DataBatch into 4 objects (let's say the inverse of CollateFunction) in the data loader.

Is there any native support? thanks.

[BraveDistribution]

https://github.com/pyg-team/pytorch_geometric/pull/4628/files/a83f0b005fb2a77ee243ac88aeeac4dbd5858050

unbatch functionality for tensor has been added, however the unbatch functionality for DataBatch is missing (fixing the changes in edge_indices).

[Padarn]

This seems like a reasonable thing to add. I'll create an issue out of this discussion. However, it would be fairly easy to manually adjust the edge_index manually for now.

Curious why you want to unbatch?

[BraveDistribution]

I use pytorch_geometric for generative task / motion prediction tasks. I need to visualize the generated graphs and compare it with the ground truth. They are 3d meshes.

I use a function that was implemented, but not merged #1574 (they use tetras as additional parameter in graphs, which is not affected by the collater function, however this functionality might be useful for all the other graphs).

[Padarn]

thanks, I think its worth getting this merged, I opened a new issue here: #4717

[rusty1s]

Isn't this already supported via batch[i] or batch.to_data_list()?

[Padarn]

(I hadn't looked into this, so please close the issue linked if this is the case @BraveDistribution @rusty1s )