Re-mapping bipartite graph connections after simplifying one of the sub graphs.

[dhorka]

Hi,

I would like to ask you if you could help me to improve a problem that I have. Let me first setup the experiment.
I have a bigraph with their relations defined in edge_index_bigraph = [indx_g1, indx_g2] where g1 is the first isolated graph and g2 is the second isolated graph. In this specific case, g1 is a point cloud. What I did is to sample the point cloud g1 to a simplified version of it, lets called this simplified version g1_simpl. To relate the g1 and g1_simpl I used the knn method of py_geo with k=1. Then what I get is a indx tensor that relates g1and g1_simpl let's called g12g1simpl

Now what I would like to do is create a new edge_index_bigraph = [indx_g1_simpl, indx_g2] which is based on the original connections of edge_index_bigraph = [indx_g1, indx_g2]. Basically what I need to do is to map and filter the old indexes with the new ones using the tensor g12g1simpl. Right now, the only way that I found to do that is using a for loop, however, this is quite slow. Do you know a better solution to solve that mapping problem that could be more efficient than using a classical loop approach?

Thanks!!

[rusty1s]

You should be able to do this via a simple mapping vector and some fancy indexing/masking logic:

mapping = torch.full((indx_g1.size(0), ), -1, dtype=torch.long)
mapping[g1_simpl] = torch.arange(g1_simple.size(0))

indx_g1 = mapping[indx_g1]
mask = indx_g1 != -1
indx_g1 = indx_g1[mask]