def read_heterograph_dgl(raw_dir,
add_inverse_edge=False,
additional_node_files=[],
additional_edge_files=[],
binary=False):
if binary:
# npz
graph_list = read_binary_heterograph_raw(raw_dir, add_inverse_edge)
else:
# csv
graph_list = read_csv_heterograph_raw(
raw_dir,
add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)
dgl_graph_list = []
print('Converting graphs into DGL objects...')
for graph in tqdm(graph_list):
g_dict = {}
# add edge connectivity
for triplet, edge_index in graph['edge_index_dict'].items():
edge_tuple = [(i, j)
for i, j in zip(graph['edge_index_dict'][triplet][0],
graph['edge_index_dict'][triplet][1])
]
g_dict[triplet] = edge_tuple
dgl_hetero_graph = dgl.heterograph(
g_dict, num_nodes_dict=graph['num_nodes_dict'])
if graph['edge_feat_dict'] is not None:
for triplet in graph['edge_feat_dict'].keys():
dgl_hetero_graph.edges[triplet].data[
'feat'] = torch.from_numpy(
graph['edge_feat_dict'][triplet])
if graph['node_feat_dict'] is not None:
for nodetype in graph['node_feat_dict'].keys():
dgl_hetero_graph.nodes[nodetype].data[
'feat'] = torch.from_numpy(
graph['node_feat_dict'][nodetype])
for key in additional_node_files:
for nodetype in graph[key].keys():
dgl_hetero_graph.nodes[nodetype].data[
key[5:]] = torch.from_numpy(graph[key][nodetype])
for key in additional_edge_files:
for triplet in graph[key].keys():
dgl_hetero_graph.edges[triplet].data[
key[5:]] = torch.from_numpy(graph[key][triplet])
dgl_graph_list.append(dgl_hetero_graph)
return dgl_graph_list
def read_heterograph_pyg(raw_dir,
add_inverse_edge=False,
additional_node_files=[],
additional_edge_files=[],
binary=False):
if binary:
# npz
graph_list = read_binary_heterograph_raw(raw_dir, add_inverse_edge)
else:
# csv
graph_list = read_csv_heterograph_raw(
raw_dir,
add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)
pyg_graph_list = []
print('Converting graphs into PyG objects...')
for graph in tqdm(graph_list):
g = Data()
g.__num_nodes__ = graph['num_nodes_dict']
g.num_nodes_dict = graph['num_nodes_dict']
# add edge connectivity
g.edge_index_dict = {}
for triplet, edge_index in graph['edge_index_dict'].items():
g.edge_index_dict[triplet] = torch.from_numpy(edge_index)
del graph['edge_index_dict']
if graph['edge_feat_dict'] is not None:
g.edge_attr_dict = {}
for triplet in graph['edge_feat_dict'].keys():
g.edge_attr_dict[triplet] = torch.from_numpy(
graph['edge_feat_dict'][triplet])
del graph['edge_feat_dict']
if graph['node_feat_dict'] is not None:
g.x_dict = {}
for nodetype in graph['node_feat_dict'].keys():
g.x_dict[nodetype] = torch.from_numpy(
graph['node_feat_dict'][nodetype])
del graph['node_feat_dict']
for key in additional_node_files:
g[key] = {}
for nodetype in graph[key].keys():
g[key][nodetype] = torch.from_numpy(graph[key][nodetype])
del graph[key]
for key in additional_edge_files:
g[key] = {}
for triplet in graph[key].keys():
g[key][triplet] = torch.from_numpy(graph[key][triplet])
del graph[key]
pyg_graph_list.append(g)
return pyg_graph_list
def pre_process(self):
processed_dir = osp.join(self.root, 'processed')
pre_processed_file_path = osp.join(processed_dir, 'data_processed')
if osp.exists(pre_processed_file_path):
# self.graph = torch.load(pre_processed_file_path, 'rb')
self.graph = load_pickle(pre_processed_file_path)
else:
### check download
if self.binary:
# npz format
has_necessary_file_simple = osp.exists(
osp.join(self.root, 'raw',
'data.npz')) and (not self.is_hetero)
has_necessary_file_hetero = osp.exists(
osp.join(self.root, 'raw',
'edge_index_dict.npz')) and self.is_hetero
else:
# csv file
has_necessary_file_simple = osp.exists(
osp.join(self.root, 'raw',
'edge.csv.gz')) and (not self.is_hetero)
has_necessary_file_hetero = osp.exists(
osp.join(self.root, 'raw',
'triplet-type-list.csv.gz')) and self.is_hetero
has_necessary_file = has_necessary_file_simple or has_necessary_file_hetero
if not has_necessary_file:
url = self.meta_info['url']
if decide_download(url):
path = download_url(url, self.original_root)
extract_zip(path, self.original_root)
os.unlink(path)
# delete folder if there exists
try:
shutil.rmtree(self.root)
except:
pass
shutil.move(
osp.join(self.original_root, self.download_name),
self.root)
else:
print('Stop download.')
exit(-1)
raw_dir = osp.join(self.root, 'raw')
### pre-process and save
add_inverse_edge = self.meta_info['add_inverse_edge'] == 'True'
if self.meta_info['additional node files'] == 'None':
additional_node_files = []
else:
additional_node_files = self.meta_info[
'additional node files'].split(',')
if self.meta_info['additional edge files'] == 'None':
additional_edge_files = []
else:
additional_edge_files = self.meta_info[
'additional edge files'].split(',')
if self.is_hetero:
if self.binary:
self.graph = read_binary_heterograph_raw(
raw_dir, add_inverse_edge=add_inverse_edge)[
0] # only a single graph
else:
self.graph = read_csv_heterograph_raw(
raw_dir,
add_inverse_edge=add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)[
0] # only a single graph
else:
if self.binary:
self.graph = read_binary_graph_raw(
raw_dir, add_inverse_edge=add_inverse_edge)[
0] # only a single graph
else:
self.graph = read_csv_graph_raw(
raw_dir,
add_inverse_edge=add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)[
0] # only a single graph
print('Saving...')
# torch.save(self.graph, pre_processed_file_path, pickle_protocol=4)
dump_pickle(self.graph, pre_processed_file_path)
def pre_process(self):
processed_dir = osp.join(self.root, 'processed')
pre_processed_file_path = osp.join(processed_dir, 'data_processed')
if osp.exists(pre_processed_file_path):
# loaded_dict = torch.load(pre_processed_file_path)
loaded_dict = load_pickle(pre_processed_file_path)
self.graph, self.labels = loaded_dict['graph'], loaded_dict[
'labels']
else:
### check download
if self.binary:
# npz format
has_necessary_file_simple = osp.exists(
osp.join(self.root, 'raw',
'data.npz')) and (not self.is_hetero)
has_necessary_file_hetero = osp.exists(
osp.join(self.root, 'raw',
'edge_index_dict.npz')) and self.is_hetero
else:
# csv file
has_necessary_file_simple = osp.exists(
osp.join(self.root, 'raw',
'edge.csv.gz')) and (not self.is_hetero)
has_necessary_file_hetero = osp.exists(
osp.join(self.root, 'raw',
'triplet-type-list.csv.gz')) and self.is_hetero
has_necessary_file = has_necessary_file_simple or has_necessary_file_hetero
if not has_necessary_file:
url = self.meta_info['url']
if decide_download(url):
path = download_url(url, self.original_root)
extract_zip(path, self.original_root)
os.unlink(path)
# delete folder if there exists
try:
shutil.rmtree(self.root)
except:
pass
shutil.move(
osp.join(self.original_root, self.download_name),
self.root)
else:
print('Stop download.')
exit(-1)
raw_dir = osp.join(self.root, 'raw')
### pre-process and save
add_inverse_edge = self.meta_info['add_inverse_edge'] == 'True'
if self.meta_info['additional node files'] == 'None':
additional_node_files = []
else:
additional_node_files = self.meta_info[
'additional node files'].split(',')
if self.meta_info['additional edge files'] == 'None':
additional_edge_files = []
else:
additional_edge_files = self.meta_info[
'additional edge files'].split(',')
if self.is_hetero:
if self.binary:
self.graph = read_binary_heterograph_raw(
raw_dir, add_inverse_edge=add_inverse_edge)[
0] # only a single graph
tmp = np.load(osp.join(raw_dir, 'node-label.npz'))
self.labels = {}
for key in list(tmp.keys()):
self.labels[key] = tmp[key]
del tmp
else:
self.graph = read_csv_heterograph_raw(
raw_dir,
add_inverse_edge=add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)[
0] # only a single graph
self.labels = read_node_label_hetero(raw_dir)
else:
if self.binary:
self.graph = read_binary_graph_raw(
raw_dir, add_inverse_edge=add_inverse_edge)[
0] # only a single graph
self.labels = np.load(osp.join(
raw_dir, 'node-label.npz'))['node_label']
else:
self.graph = read_csv_graph_raw(
raw_dir,
add_inverse_edge=add_inverse_edge,
additional_node_files=additional_node_files,
additional_edge_files=additional_edge_files)[
0] # only a single graph
self.labels = pd.read_csv(osp.join(raw_dir,
'node-label.csv.gz'),
compression='gzip',
header=None).values
print('Saving...')
# torch.save({'graph': self.graph, 'labels': self.labels}, pre_processed_file_path, pickle_protocol=4)
dump_pickle({
'graph': self.graph,
'labels': self.labels
}, pre_processed_file_path)
def _save_graph_list_hetero(self, graph_list):
dict_keys = graph_list[0].keys()
# check necessary keys
if not 'edge_index_dict' in dict_keys:
raise RuntimeError(
'edge_index_dict needs to be provided in graph objects')
if not 'num_nodes_dict' in dict_keys:
raise RuntimeError(
'num_nodes_dict needs to be provided in graph objects')
print(dict_keys)
# Store the following files
# - edge_index_dict.npz (necessary)
# edge_index_dict
# - num_nodes_dict.npz (necessary)
# num_nodes_dict
# - num_edges_dict.npz (necessary)
# num_edges_dict
# - node_**.npz (optional, node_feat_dict is the default node features)
# - edge_**.npz (optional, edge_feat_dict the default edge features)
# extract entity types
ent_type_list = sorted(
[e for e in graph_list[0]['num_nodes_dict'].keys()])
# saving num_nodes_dict
print('Saving num_nodes_dict')
num_nodes_dict = {}
for ent_type in ent_type_list:
num_nodes_dict[ent_type] = np.array([
graph['num_nodes_dict'][ent_type] for graph in graph_list
]).astype(np.int64)
np.savez_compressed(osp.join(self.raw_dir, 'num_nodes_dict.npz'),
**num_nodes_dict)
print(num_nodes_dict)
# extract triplet types
triplet_type_list = sorted([
(h, r, t) for (h, r, t) in graph_list[0]['edge_index_dict'].keys()
])
print(triplet_type_list)
# saving edge_index_dict
print('Saving edge_index_dict')
num_edges_dict = {}
edge_index_dict = {}
for triplet in triplet_type_list:
# representing triplet (head, rel, tail) as a single string 'head___rel___tail'
triplet_cat = '___'.join(triplet)
edge_index = np.concatenate(
[graph['edge_index_dict'][triplet] for graph in graph_list],
axis=1).astype(np.int64)
if edge_index.shape[0] != 2:
raise RuntimeError('edge_index must have shape (2, num_edges)')
num_edges = np.array([
graph['edge_index_dict'][triplet].shape[1]
for graph in graph_list
]).astype(np.int64)
num_edges_dict[triplet_cat] = num_edges
edge_index_dict[triplet_cat] = edge_index
print(edge_index_dict)
print(num_edges_dict)
np.savez_compressed(osp.join(self.raw_dir, 'edge_index_dict.npz'),
**edge_index_dict)
np.savez_compressed(osp.join(self.raw_dir, 'num_edges_dict.npz'),
**num_edges_dict)
for key in dict_keys:
if key == 'edge_index_dict' or key == 'num_nodes_dict':
continue
if graph_list[0][key] is None:
continue
print(f'Saving {key}')
feat_dict = {}
if 'node_' in key:
# node feature dictionary
for ent_type in graph_list[0][key].keys():
if ent_type not in num_nodes_dict:
raise RuntimeError(
f'Encountered unknown entity type called {ent_type}.'
)
# check num_nodes
for i in range(len(graph_list)):
if len(graph_list[i][key]
[ent_type]) != num_nodes_dict[ent_type][i]:
raise RuntimeError(
f'num_nodes mistmatches with {key}[{ent_type}]'
)
# make sure saved in np.int64 or np.float32
dtype = np.int64 if 'int' in str(
graph_list[0][key][ent_type].dtype) else np.float32
cat_feat = np.concatenate(
[graph[key][ent_type] for graph in graph_list],
axis=0).astype(dtype)
feat_dict[ent_type] = cat_feat
elif 'edge_' in key:
# edge feature dictionary
for triplet in graph_list[0][key].keys():
# representing triplet (head, rel, tail) as a single string 'head___rel___tail'
triplet_cat = '___'.join(triplet)
if triplet_cat not in num_edges_dict:
raise RuntimeError(
f"Encountered unknown triplet type called ({','.join(triplet)})."
)
# check num_edges
for i in range(len(graph_list)):
if len(graph_list[i][key]
[triplet]) != num_edges_dict[triplet_cat][i]:
raise RuntimeError(
f"num_edges mismatches with {key}[({','.join(triplet)})]"
)
# make sure saved in np.int64 or np.float32
dtype = np.int64 if 'int' in str(
graph_list[0][key][triplet].dtype) else np.float32
cat_feat = np.concatenate(
[graph[key][triplet] for graph in graph_list],
axis=0).astype(dtype)
feat_dict[triplet_cat] = cat_feat
else:
raise RuntimeError(
f'Keys in graph object should start from either \'node_\' or \'edge_\', but \'{key}\' given.'
)
np.savez_compressed(osp.join(self.raw_dir, f'{key}.npz'),
**feat_dict)
print('Validating...')
# testing
print('Reading saved files')
graph_list_read = read_binary_heterograph_raw(self.raw_dir, False)
print('Checking read graphs and given graphs are the same')
for i in tqdm(range(len(graph_list))):
for key0, value0 in graph_list[i].items():
if value0 is not None:
for key1, value1 in value0.items():
if isinstance(graph_list[i][key0][key1], np.ndarray):
assert (np.allclose(graph_list[i][key0][key1],
graph_list_read[i][key0][key1],
rtol=1e-04,
atol=1e-04,
equal_nan=True))
else:
assert (graph_list[i][key0][key1] ==
graph_list_read[i][key0][key1])
del graph_list_read
