The gipa_wide_deep from houyili

gipa_wide_deep's People

Contributors

Stargazers

Watchers

gipa_wide_deep's Issues

some question about trans_edge_fea_to_sparse

hello author, I am a beginner in GNN and have some questions about GIPA code implementation. Can you explain the process for transforming sparse tensors, how do you get interval list parameters? thanks a lot :)

interval_0_1 = [0.001, 1]
interval_3 = [0.001, 0.7, 1]
interval_4 = [0.001, 0.1, 0.2, 1]
interval_12 = [0.001, 0.07, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.5, 0.7, 1]
interval_15 = [0.001, 0.07, 0.1, 0.15, 0.2, 0.25, 0.3, 0.35, 0.4, 0.45, 0.5, 0.6, 0.7, 0.8, 1]
simple_inter = [interval_0_1, interval_4, interval_0_1, interval_4, interval_12, interval_12, interval_3, interval_15]


def trans_edge_fea_to_sparse(raw_edge_fea, graph, interval: list, is_log=False):
    edge_fea_list = []
    for i in range(8):
        print("Process edge feature == %d " % i)
        res = torch.reshape((raw_edge_fea[:, i] == 0.001).float(), [-1, 1])
        edge_fea_list.append(res)
        for j in range(1, len(interval[i])):
            small, big = float(interval[i][j - 1]), float(interval[i][j])
            print("process interval %0.3f < x <= %0.3f " % (small, big))
            cond = torch.logical_and((raw_edge_fea[:, i] > small), (raw_edge_fea[:, i] <= big))
            edge_fea_list.append(torch.reshape(cond.float(), [-1, 1]))
    sparse = torch.concat(edge_fea_list, dim=-1)
    print(sparse.size())
    graph.edata.update({"sparse": sparse})
    graph.update_all(fn.copy_e("sparse", "sparse_c"), fn.sum("sparse_c", "sparse_f" if is_log else "sparse"))
    if is_log:
        graph.apply_nodes(lambda nodes: {"sparse": torch.log2(nodes.data['sparse_f'] + 1)})
        del graph.ndata["sparse_f"]
    return sparse