Apply node-level MSE on batches of multiple graphs

[bardia-esmaeili]

Hello,

I am trying to measure the below loss function for a batch of multiple graphs with a different number of nodes for each graph.

As you can see in the formula, I should measure the mse for each node embedding in each graph, measure the mean over that graph, then average the value for all graphs.

Since a data loader in PyG provides x, edge_index, y, batch, ptr, is there a way to calculate such a loss over a batch in a parallel form?

[rusty1s]

You can use scatter_mean to compute the mean per graph, and then use mean() to compute the mean per batch.

[bardia-esmaeili]

Thank you for your fast response.