Neighbor Sampling in Heterogeneous Graphs

aakanksha · July 2, 2020, 2:10pm

Hi,

I was trying to sample neighbors of nodes in a heterogeneous graph using the dgl.sampling.sample_neighbors and dgl.to_block APIs in accordance with the “sample_blocks” method of the “NeighborSampler” class defined in this example. However, I run into the following error whenever I try to proceed with extracting seeds for the next block using seeds = block.srcdata[dgl.NID]:
AssertionError: srcdata is only allowed when there is only one SRC type.

Could you please help me understand how I can possibly use these APIs to sample nodes from a heterogeneous graph?

Please let me know if there are any other methods which I could make use of for the same without converting the graph to a homogeneous graph.

Thanks.

BarclayII · July 6, 2020, 1:26am

Hi, we have a minibatch training example for RGCN that directly work by neighbor sampling on heterogeneous graphs in

github.com

dmlc/dgl/blob/master/examples/pytorch/rgcn-hetero/entity_classify_mb.py

"""Modeling Relational Data with Graph Convolutional Networks
Paper: https://arxiv.org/abs/1703.06103
Reference Code: https://github.com/tkipf/relational-gcn
"""
import argparse
import itertools
import numpy as np
import time
import torch as th
import torch.nn as nn
import torch.nn.functional as F
from torch.utils.data import DataLoader
from functools import partial

import dgl
from dgl.data.rdf import AIFB, MUTAG, BGS, AM
from model import EntityClassify, RelGraphEmbed

def extract_embed(node_embed, input_nodes):
    emb = {}

This file has been truncated. show original

Note that this require the nightly build to work.

Please feel free to follow up. Thanks.