torch_geometric.nn.models.VGAE has no provision for batch processing, right?

[survivebycoding]

This function has no provision for batch processing right?

https://pytorch-geometric.readthedocs.io/en/2.5.2/generated/torch_geometric.nn.models.VGAE.html#torch_geometric.nn.models.VGAE

If there is a way, can someone kindly help me out with it please?

[Romain-Nicolle]

Hello,

I am not an expert but I will try to help you.
So, the source code for the VGAE model is there.

Here you can find the source code for the GraphSAGE model which support batch processing. The message_passing_convolution used by the GraphSAGE model is the SAGEConv, whose source code is there.
Interestingly, even though the SAGEConv doesn't seem to support batch_processing, the GraphSAGE model can support it, because it is written that way in the forward function.

So I gues, you could do that too. Rewrite some parts of the code in order for the functions to handle batches ?

Besides, have you checkes this question related to batches and VGAE ? Maybe it could help you ? Depending on whether you want to have mini-batches of one big graph or batches of several small graphes.

Hope that helps.

Best,