How can i get the reconstructed network using bionic?

[chinaliye]

I found that the output of bionic was just node embeddings of the fused network. I wonder if there is a simple way to get the edges and weights of the fused network(like the input networks), without changing the code of bionic.

[duncster94]

Hi, the best way to do this is to compute a pairwise cosine similarity of the embeddings, followed by thresholding the resulting similarities. You can do this using radius_neighbors_graph from sklearn:

def make_net(feat: pd.DataFrame, radius: float) -> nx.Graph:
    net = radius_neighbors_graph(
        feat.values, radius, mode="connectivity", metric="cosine"
    )
    net = nx.from_scipy_sparse_array(net)
    net = nx.relabel_nodes(net, {idx: node for idx, node in enumerate(feat.index)})
    net.remove_edges_from(nx.selfloop_edges(net))
    net.remove_nodes_from(list(nx.isolates(net)))
    return net