How to construct heter-blocks when using DistHeterGraphSAGE

Hi, I’m trying to train heter-graph with GraphSAGE in distributed mode. Following the demos in ‘examples/pytorch/rgcn/experiment’ and 'examples/pytorch/graphsage/experiment ',I found that the DGL actually sampled homo-graphs and doesn’t convert homo-blocks back to heter-blocks.
I wanna use GraphSAGE with my heter-graph, so the heter-blocks is necessary in my scenario.
I read the source code and tried to write my homo2heter function, but it doesn’t work well.
Here is my codes in ‘sample_blocks’ and the error messages.

class SimpleMeta:
def edges(self, keys=True):
    # (srctype, dsttype, etype)
    return [
        ("company", "person", "company-person"),
        ("person", "company", "person-company"),
        ("person", "person", "person-person")

def sample_blocks(self, seeds):
    blocks = []
    etypes = []
    norms = []
    ntypes = []
    seeds = th.LongTensor(np.asarray(seeds))
    gpb = self.g.get_partition_book()
    # We need to map the per-type node IDs to homogeneous IDs.
    next_global_homo_seeds = gpb.map_to_homo_nid(seeds, 'person')
for fanout in self.fanouts:
    # For a heterogeneous input graph, the returned frontier is stored in
    # the homogeneous graph format.
    # -----------------------org-----------------------------
    curr_global_homo_seeds = next_global_homo_seeds
    homo_frontier = self.sample_neighbors(self.g, next_global_homo_seeds, fanout, replace=False)
    block_tmp = dgl.to_block(homo_frontier, next_global_homo_seeds)
    next_global_homo_seeds = block_tmp.srcdata[dgl.NID]
    # -----------------------org-----------------------------

    homo_frontier.ndata[dgl.NID] = th.arange(homo_frontier.number_of_nodes())
    homo_frontier.ndata[dgl.NTYPE], _ = gpb.map_to_per_ntype(homo_frontier.ndata[dgl.NID])
    homo_frontier.edata[dgl.ETYPE], _ = gpb.map_to_per_etype(homo_frontier.edata[dgl.EID])

    heter_seed_nodes = {"person": [], "company": []}
    for nid in curr_global_homo_seeds:
        nid = nid.item()
        ntype = homo_frontier.ndata[dgl.NTYPE][nid].item()
        if ntype == 0:

    heter_frontier = dgl.to_heterogeneous(homo_frontier,
                                          ntypes=["company", "person"], #self.g.ntypes
                                          etypes=["company-person", "person-company", "person-person"], #self.g.etypes
    heter_block = dgl.to_block(heter_frontier, heter_seed_nodes)
    blocks.insert(0, heter_block)
return seeds, blocks

error message

Is there any idea to convert homo-blocks to heter-blocks or why my code raise this error message?