Slicing a batch

grendelaglaeca · April 9, 2022, 10:38pm

Hi, was slice_batch API deprecated? I tried to install it from source but slice_idx module seems to be missing in git.
Is there another way to access the edges IDs of specific graphs in a batch?

minjie · April 10, 2022, 2:48am

Edges of the same graph are stored together in a batch. Suppose there are three graphs, each with N_0, N_1, N_2 edges respectively. Then when batched together, these edges are stored in [0, N_0), [N_0, N_0+N_1), [N_0+N_1~N_0+N_1+N_2) respectively.

Rhett-Ying · April 10, 2022, 2:55am

dgl.slice_batch is NOT deprecated. Here’s the doc link: dgl.slice_batch — DGL 0.8.0post2 documentation.

Which branch of source code are you using? Below is master branch.

github.com

dmlc/dgl/blob/01e50626d3be0f3fc6ade4731c3793927700da3e/python/dgl/batch.py#L419

      
        
            
            
    # Unbatch edge features
                for etype in g.canonical_etypes:
                    for key, feat in g.edges[etype].data.items():
                        subfeats = F.split(feat, edge_split[etype], 0)
                        for subg, subf in zip(gs, subfeats):
                            subg.edges[etype].data[key] = subf
            
            
    return gs
            
            
def slice_batch(g, gid, store_ids=False):
                """Get a particular graph from a batch of graphs.
            
            
    Parameters
                ----------
                g : DGLGraph
                    Input batched graph.
                gid : int
                    The ID of the graph to retrieve.
                store_ids : bool
                    If True, it will store the raw IDs of the extracted nodes and edges in the ``ndata`` and

system · May 10, 2022, 2:55am

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.