How to do alignment with PyG format xyz?

[StefanIsSmart]

When we are dealing with compounds, sometimes we need to predict the structure.
The question is how to align them with the batch.

For example, there are pred_xyz and label_xyz, we need to use SVD to align them and then use the MAEloss to calculate the loss of prediction.
In PyG, the format of pred_xyz and label_xyz with be : (N, 3).[In fact, N is the sum of nodes from G compounds (So the batch_size of compounds is G, and the batch_size of Node is N, the length of every compound is different)]

We cannot directly use SVD to align de pred_xyz and label_xyz, because that will lead to the compound_(1) will consider the coordinates of compound_(2-G) during the alignment.

I have to do that by for-loop, but it is very ineffective.

How could I do that with batch rather than for-loop?

[rusty1s]

The recommended way to do this is to convert your matrix into a 3-dimensional tensor with a batch dimension via utils.to_dense_batch. Note that you need to take care of proper padding treatment in this case.