def load_dataset(): ''' load raw datasets. :return: a list of networkx/deepsnap graphs, plus additional info if needed ''' format = cfg.dataset.format name = cfg.dataset.name # dataset_dir = '{}/{}'.format(cfg.dataset.dir, name) dataset_dir = cfg.dataset.dir # Try to load customized data format for func in register.loader_dict.values(): graphs = func(format, name, dataset_dir) if graphs is not None: return graphs # Load from Pytorch Geometric dataset if format == 'PyG': graphs = load_pyg(name, dataset_dir) # Load from networkx formatted data # todo: clean nx dataloader elif format == 'nx': graphs = load_nx(name, dataset_dir) # Load from OGB formatted data elif cfg.dataset.format == 'OGB': if cfg.dataset.name == 'ogbg-molhiv': dataset = PygGraphPropPredDataset(name=cfg.dataset.name) graphs = GraphDataset.pyg_to_graphs(dataset) # Note this is only used for custom splits from OGB split_idx = dataset.get_idx_split() return graphs, split_idx else: raise ValueError('Unknown data format: {}'.format(cfg.dataset.format)) return graphs
def test_resample_disjoint(self): pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) graph = graphs[0] graph = Graph(node_label=graph.node_label, node_feature=graph.node_feature, edge_index=graph.edge_index, edge_feature=graph.edge_feature, directed=False) graphs = [graph] dataset = GraphDataset(graphs, task="link_pred", edge_train_mode="disjoint", edge_message_ratio=0.8, resample_disjoint=True, resample_disjoint_period=1) dataset_train, _, _ = dataset.split(split_ratio=[0.5, 0.2, 0.3]) graph_train_first = dataset_train[0] graph_train_second = dataset_train[0] self.assertEqual(graph_train_first.edge_label_index.shape[1], graph_train_second.edge_label_index.shape[1]) self.assertTrue( torch.equal(graph_train_first.edge_label, graph_train_second.edge_label))
def train(rank, pygds, args, num_node_features, num_classes): if args.skip is not None: model_cls = skip_models.SkipLastGNN elif args.model == "GIN": model_cls = GIN else: model_cls = GNN model = model_cls(num_node_features, args.hidden_dim, num_classes, args).to(device) opt = build_optimizer(args, model.parameters()) # DISTRIBUTED TRAINING - can be replaced with 1 call to model_parallelize() world_size = torch.cuda.device_count() os.environ['MASTER_ADDR'] = 'localhost' os.environ['MASTER_PORT'] = '12355' pyg_dataset[0] = pyg_dataset[0].split(pyg_dataset[0].size(0) // world_size)[rank] device = rank model = DistributedDataParallel(model) graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="graph") datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio = [0.8, 0.1, 0.1])
def test_torch_dataloader_collate(self): # graph classification example pyg_dataset = TUDataset('./enzymes', 'ENZYMES') graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="graph") train_batch_num = math.ceil(len(dataset) * 0.8 / 32) test_batch_num = math.ceil(len(dataset) * 0.1 / 32) val_batch_num = math.ceil(len(dataset) * 0.1 / 32) datasets = {} datasets['train'], datasets['val'], datasets['test'] = \ dataset.split(transductive=False, split_ratio=[0.8, 0.1, 0.1]) dataloaders = { split: DataLoader(dataset, collate_fn=Batch.collate(), batch_size=32, shuffle=True) for split, dataset in datasets.items() } self.assertEqual(len(dataloaders['train']), train_batch_num) self.assertEqual(len(dataloaders['val']), test_batch_num) self.assertEqual(len(dataloaders['test']), val_batch_num) for i, data in enumerate(dataloaders['train']): if i != len(dataloaders['train']) - 1: self.assertEqual(data.num_graphs, 32) for i, data in enumerate(dataloaders['val']): if i != len(dataloaders['val']) - 1: self.assertEqual(data.num_graphs, 32) for i, data in enumerate(dataloaders['test']): if i != len(dataloaders['test']) - 1: self.assertEqual(data.num_graphs, 32)
def load_dataset_example(format, name, dataset_dir): dataset_dir = '{}/{}'.format(dataset_dir, name) if format == 'PyG': if name == 'QM7b': dataset_raw = QM7b(dataset_dir) graphs = GraphDataset.pyg_to_graphs(dataset_raw) return graphs
def deepsnap_ego(args, pyg_dataset): avg_time = 0 task = "graph" for i in range(args.num_runs): if args.print_run: print("Run {}".format(i + 1)) graphs = GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, netlib=netlib) dataset = GraphDataset(graphs, task=task) datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio=[0.8, 0.1, 0.1], shuffle=False) dataloaders = { split: DataLoader(dataset, collate_fn=Batch.collate(), batch_size=1, shuffle=False) for split, dataset in datasets.items() } s = time.time() for batch in dataloaders['train']: batch = batch.apply_transform(ego_nets, update_tensor=True) avg_time += (time.time() - s) print("DeepSNAP has average time: {}".format(avg_time / args.num_runs))
def test_pyg_to_graphs_global(self): import deepsnap deepsnap.use(nx) pyg_dataset = Planetoid('./planetoid', "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) self.assertTrue(isinstance(graphs[0].G, nx.Graph)) dataset = GraphDataset(graphs, task='node') num_nodes = dataset.num_nodes[0] node_0 = int(0.8 * num_nodes) node_1 = int(0.1 * num_nodes) node_2 = num_nodes - node_0 - node_1 train, val, test = dataset.split() self.assertTrue(isinstance(train[0].G, nx.Graph)) self.assertTrue(isinstance(val[0].G, nx.Graph)) self.assertTrue(isinstance(test[0].G, nx.Graph)) self.assertEqual(train[0].node_label_index.shape[0], node_0) self.assertEqual(val[0].node_label_index.shape[0], node_1) self.assertEqual(test[0].node_label_index.shape[0], node_2) train_loader = DataLoader(train, collate_fn=Batch.collate(), batch_size=1) for batch in train_loader: self.assertTrue(isinstance(batch.G[0], nx.Graph)) deepsnap.use(sx) graphs = GraphDataset.pyg_to_graphs(pyg_dataset) self.assertTrue(isinstance(graphs[0].G, sx.Graph)) dataset = GraphDataset(graphs, task='node') num_nodes = dataset.num_nodes[0] node_0 = int(0.8 * num_nodes) node_1 = int(0.1 * num_nodes) node_2 = num_nodes - node_0 - node_1 train, val, test = dataset.split() self.assertTrue(isinstance(train[0].G, sx.Graph)) self.assertTrue(isinstance(val[0].G, sx.classes.graph.Graph)) self.assertTrue(isinstance(test[0].G, sx.classes.graph.Graph)) self.assertEqual(train[0].node_label_index.shape[0], node_0) self.assertEqual(val[0].node_label_index.shape[0], node_1) self.assertEqual(test[0].node_label_index.shape[0], node_2) train_loader = DataLoader(train, collate_fn=Batch.collate(), batch_size=1) for batch in train_loader: self.assertTrue(isinstance(batch.G[0], sx.Graph))
def main(): args = arg_parse() pyg_dataset = Planetoid('./cora', 'Cora', transform=T.TargetIndegree()) # the input that we assume users have edge_train_mode = args.mode print('edge train mode: {}'.format(edge_train_mode)) graphs = GraphDataset.pyg_to_graphs(pyg_dataset, tensor_backend=True) if args.multigraph: graphs = [copy.deepcopy(graphs[0]) for _ in range(10)] dataset = GraphDataset(graphs, task='link_pred', edge_message_ratio=args.edge_message_ratio, edge_train_mode=edge_train_mode) print('Initial dataset: {}'.format(dataset)) # split dataset datasets = {} datasets['train'], datasets['val'], datasets['test']= dataset.split( transductive=not args.multigraph, split_ratio=[0.85, 0.05, 0.1]) print('after split') print('Train message-passing graph: {} nodes; {} edges.'.format( datasets['train'][0].num_nodes, datasets['train'][0].num_edges)) print('Val message-passing graph: {} nodes; {} edges.'.format( datasets['val'][0].num_nodes, datasets['val'][0].num_edges)) print('Test message-passing graph: {} nodes; {} edges.'.format( datasets['test'][0].num_nodes, datasets['test'][0].num_edges)) # node feature dimension input_dim = datasets['train'].num_node_features # link prediction needs 2 classes (0, 1) num_classes = datasets['train'].num_edge_labels model = Net(input_dim, num_classes, args).to(args.device) #optimizer = torch.optim.Adam(model.parameters(), lr=0.001, weight_decay=5e-3) optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.9, weight_decay=5e-4) scheduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=args.epochs) follow_batch = [] # e.g., follow_batch = ['edge_index'] dataloaders = {split: DataLoader( ds, collate_fn=Batch.collate(follow_batch), batch_size=args.batch_size, shuffle=(split=='train')) for split, ds in datasets.items()} print('Graphs after split: ') for key, dataloader in dataloaders.items(): for batch in dataloader: print(key, ': ', batch) train(model, dataloaders, optimizer, args, scheduler=scheduler)
def load_pyg(name, dataset_dir): ''' load pyg format dataset :param name: dataset name :param dataset_dir: data directory :return: a list of networkx/deepsnap graphs ''' dataset_dir = '{}/{}'.format(dataset_dir, name) if name in ['Cora', 'CiteSeer', 'PubMed']: dataset_raw = Planetoid(dataset_dir, name) elif name[:3] == 'TU_': # TU_IMDB doesn't have node features if name[3:] == 'IMDB': name = 'IMDB-MULTI' dataset_raw = TUDataset(dataset_dir, name, transform=T.Constant()) else: dataset_raw = TUDataset(dataset_dir, name[3:]) # TU_dataset only has graph-level label # The goal is to have synthetic tasks # that select smallest 100 graphs that have more than 200 edges if cfg.dataset.tu_simple and cfg.dataset.task != 'graph': size = [] for data in dataset_raw: edge_num = data.edge_index.shape[1] edge_num = 9999 if edge_num < 200 else edge_num size.append(edge_num) size = torch.tensor(size) order = torch.argsort(size)[:100] dataset_raw = dataset_raw[order] elif name == 'Karate': dataset_raw = KarateClub() elif 'Coauthor' in name: if 'CS' in name: dataset_raw = Coauthor(dataset_dir, name='CS') else: dataset_raw = Coauthor(dataset_dir, name='Physics') elif 'Amazon' in name: if 'Computers' in name: dataset_raw = Amazon(dataset_dir, name='Computers') else: dataset_raw = Amazon(dataset_dir, name='Photo') elif name == 'MNIST': dataset_raw = MNISTSuperpixels(dataset_dir) elif name == 'PPI': dataset_raw = PPI(dataset_dir) elif name == 'QM7b': dataset_raw = QM7b(dataset_dir) else: raise ValueError('{} not support'.format(name)) graphs = GraphDataset.pyg_to_graphs(dataset_raw) return graphs
def deepsnap_pagerank(args, pyg_dataset): avg_time = 0 task = 'graph' for i in range(args.num_runs): if args.print_run: print("Run {}".format(i + 1)) graphs = GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=False, netlib=netlib) dataset = GraphDataset(graphs, task=task) s = time.time() dataset.apply_transform(page_fun, update_tensor=False, lib=args.netlib) avg_time += (time.time() - s) print("DeepSNAP has average time: {}".format(avg_time / args.num_runs))
def load_dataset(name): task = "graph" if name == "enzymes": dataset = TUDataset(root="/tmp/ENZYMES", name="ENZYMES") elif name == "cox2": dataset = TUDataset(root="/tmp/cox2", name="COX2") elif name == "imdb-binary": dataset = TUDataset(root="/tmp/IMDB-BINARY", name="IMDB-BINARY") if task == "graph": dataset = GraphDataset(GraphDataset.pyg_to_graphs(dataset)) dataset = dataset.apply_transform( lambda g: g.G.subgraph(max(nx.connected_components(g.G), key=len))) dataset = dataset.filter(lambda g: len(g.G) >= 6) train, test = dataset.split(split_ratio=[0.8, 0.2]) return train, test, task
def test_filter(self): pyg_dataset = TUDataset('./enzymes', 'ENZYMES') graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="graph") thresh = 90 orig_dataset_size = len(dataset) num_graphs_large = 0 for graph in dataset: if len(graph.G) >= thresh: num_graphs_large += 1 dataset = dataset.filter( lambda graph: len(graph.G) < thresh, deep_copy=False) filtered_dataset_size = len(dataset) self.assertEqual( orig_dataset_size - filtered_dataset_size, num_graphs_large)
def load_dataset(name): def add_feats(graph): for v in graph.G.nodes: graph.G.nodes[v]["node_feature"] = torch.ones(1) return graph task = "graph" if name == "enzymes": dataset = TUDataset(root="/tmp/ENZYMES", name="ENZYMES") elif name == "cox2": dataset = TUDataset(root="/tmp/cox2", name="COX2") elif name == "imdb-binary": dataset = TUDataset(root="/tmp/IMDB-BINARY", name="IMDB-BINARY") if task == "graph": dataset = GraphDataset(GraphDataset.pyg_to_graphs(dataset)) # add blank features for imdb-binary, which doesn't have node labels if name == "imdb-binary": dataset = dataset.apply_transform(add_feats) dataset = dataset.apply_transform( lambda g: g.G.subgraph(max(nx.connected_components(g.G), key=len))) dataset = dataset.filter(lambda g: len(g.G) >= 6) train, test = dataset.split(split_ratio=[0.8, 0.2]) return train, test, task
raise ValueError("Unsupported dataset.") if args.netlib == "nx": import networkx as netlib print("Use NetworkX as the backend network library.") elif args.netlib == "sx": import snap import snapx as netlib print("Use SnapX as the backend network library.") else: raise ValueError("{} network library is not supported.".format( args.netlib)) if args.split == 'random': graphs = GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=False, netlib=netlib) dataset = GraphDataset(graphs, task='node') # node, edge, link_pred, graph dataset_train, dataset_val, dataset_test = dataset.split( transductive=True, split_ratio=[0.8, 0.1, 0.1]) # transductive split, inductive split else: graphs_train, graphs_val, graphs_test = \ GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=True, netlib=netlib) dataset_train, dataset_val, dataset_test = \ GraphDataset(graphs_train, task='node'), GraphDataset(graphs_val,task='node'), \ GraphDataset(graphs_test, task='node')
def test_dataset_split_custom(self): # transductive split with node task (self defined dataset) G, x, y, edge_x, edge_y, edge_index, graph_x, graph_y = ( simple_networkx_graph_alphabet()) Graph.add_edge_attr(G, "edge_feature", edge_x) Graph.add_edge_attr(G, "edge_label", edge_y) Graph.add_node_attr(G, "node_feature", x) Graph.add_node_attr(G, "node_label", y) Graph.add_graph_attr(G, "graph_feature", graph_x) Graph.add_graph_attr(G, "graph_label", graph_y) num_nodes = len(list(G.nodes)) nodes_train = list(G.nodes)[:int(0.3 * num_nodes)] nodes_val = list(G.nodes)[int(0.3 * num_nodes):int(0.6 * num_nodes)] nodes_test = list(G.nodes)[int(0.6 * num_nodes):] graph = Graph(G, custom_splits=[nodes_train, nodes_val, nodes_test], task="node") graphs = [graph] dataset = GraphDataset( graphs, task="node", general_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].node_label_index, list(range(int(0.3 * num_nodes)))) self.assertEqual( split_res[1][0].node_label_index, list(range(int(0.3 * num_nodes), int(0.6 * num_nodes)))) self.assertEqual(split_res[2][0].node_label_index, list(range(int(0.6 * num_nodes), num_nodes))) # transductive split with link_pred task (disjoint mode) (self defined dataset) edges = list(G.edges) num_edges = len(edges) edges_train = edges[:int(0.3 * num_edges)] edges_train_disjoint = edges[:int(0.5 * 0.3 * num_edges)] edges_val = edges[int(0.3 * num_edges):int(0.6 * num_edges)] edges_test = edges[int(0.6 * num_edges):] link_size_list = [ len(edges_train_disjoint), len(edges_val), len(edges_test) ] graph = Graph(G, custom_splits=[edges_train, edges_val, edges_test], custom_disjoint_split=edges_train_disjoint, task="link_pred") graphs = [graph] dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", general_split_mode="custom", disjoint_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * link_size_list[1]) self.assertEqual(split_res[2][0].edge_label_index.shape[1], 2 * link_size_list[2]) # transductive split with link_pred task (disjoint mode) (self defined disjoint data) edges = list(G.edges) num_edges = len(edges) edges_train = edges[:int(0.7 * num_edges)] edges_train_disjoint = edges[:int(0.5 * 0.7 * num_edges)] edges_val = edges[int(0.7 * num_edges):] link_size_list = [len(edges_train_disjoint), len(edges_val)] graph = Graph(G, custom_splits=[ edges_train, edges_val, ], custom_disjoint_split=edges_train_disjoint, task="link_pred") graphs = [graph] dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", general_split_mode="custom", disjoint_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * link_size_list[1]) # transductive split with link_pred task (disjoint mode) (self defined disjoint data) (multigraph) (train/val split) G, x, y, edge_x, edge_y, edge_index, graph_x, graph_y = ( simple_networkx_multigraph()) Graph.add_edge_attr(G, "edge_feature", edge_x) Graph.add_edge_attr(G, "edge_label", edge_y) Graph.add_node_attr(G, "node_feature", x) Graph.add_node_attr(G, "node_label", y) Graph.add_graph_attr(G, "graph_feature", graph_x) Graph.add_graph_attr(G, "graph_label", graph_y) edges = list(G.edges) num_edges = len(edges) edges_train = edges[:int(0.6 * num_edges)] edges_train_disjoint = edges[:int(0.6 * 0.2 * num_edges)] edges_val = edges[int(0.6 * num_edges):] link_size_list = [len(edges_train_disjoint), len(edges_val)] graph = Graph(G, custom_splits=[ edges_train, edges_val, ], custom_disjoint_split=edges_train_disjoint, task="link_pred") graphs = [graph] dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", general_split_mode="custom", disjoint_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * link_size_list[1]) # transductive split with link_pred task (disjoint mode) (self defined disjoint data) (multigraph) (train/val/test split) G, x, y, edge_x, edge_y, edge_index, graph_x, graph_y = ( simple_networkx_multigraph()) Graph.add_edge_attr(G, "edge_feature", edge_x) Graph.add_edge_attr(G, "edge_label", edge_y) Graph.add_node_attr(G, "node_feature", x) Graph.add_node_attr(G, "node_label", y) Graph.add_graph_attr(G, "graph_feature", graph_x) Graph.add_graph_attr(G, "graph_label", graph_y) edges = list(G.edges) num_edges = len(edges) edges_train = edges[:int(0.6 * num_edges)] edges_train_disjoint = edges[:int(0.6 * 0.2 * num_edges)] edges_val = edges[int(0.6 * num_edges):int(0.8 * num_edges)] edges_test = edges[int(0.8 * num_edges):] link_size_list = [ len(edges_train_disjoint), len(edges_val), len(edges_test) ] graph = Graph(G, custom_splits=[ edges_train, edges_val, edges_test, ], custom_disjoint_split=edges_train_disjoint, task="link_pred") graphs = [graph] dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", general_split_mode="custom", disjoint_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * link_size_list[1]) self.assertEqual(split_res[2][0].edge_label_index.shape[1], 2 * link_size_list[2]) # transductive split with node task (pytorch geometric dataset) pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] node_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: custom_splits = [[] for i in range(len(split_ratio))] split_offset = 0 shuffled_node_indices = torch.randperm(graph.num_nodes) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = (1 + int(split_ratio_i * (graph.num_nodes - len(split_ratio)))) nodes_split_i = ( shuffled_node_indices[split_offset:split_offset + num_split_i]) split_offset += num_split_i else: nodes_split_i = shuffled_node_indices[split_offset:] custom_splits[i] = nodes_split_i node_size_list[i] += len(nodes_split_i) graph.custom_splits = custom_splits dataset = GraphDataset( graphs, task="node", general_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(len(split_res[0][0].node_label_index), node_size_list[0]) self.assertEqual(len(split_res[1][0].node_label_index), node_size_list[1]) self.assertEqual(len(split_res[2][0].node_label_index), node_size_list[2]) # transductive split with edge task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] edge_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: custom_splits = [[] for i in range(len(split_ratio))] split_offset = 0 edges = list(graph.G.edges) random.shuffle(edges) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = (1 + int(split_ratio_i * (graph.num_edges - len(split_ratio)))) edges_split_i = (edges[split_offset:split_offset + num_split_i]) split_offset += num_split_i else: edges_split_i = edges[split_offset:] custom_splits[i] = edges_split_i edge_size_list[i] += len(edges_split_i) graph.custom_splits = custom_splits dataset = GraphDataset( graphs, task="edge", general_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * edge_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * edge_size_list[1]) self.assertEqual(split_res[2][0].edge_label_index.shape[1], 2 * edge_size_list[2]) # transductive split with link_pred task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] link_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: split_offset = 0 edges = list(graph.G.edges) random.shuffle(edges) num_edges_train = 1 + int(split_ratio[0] * (graph.num_edges - 3)) num_edges_val = 1 + int(split_ratio[0] * (graph.num_edges - 3)) edges_train = edges[:num_edges_train] edges_val = edges[num_edges_train:num_edges_train + num_edges_val] edges_test = edges[num_edges_train + num_edges_val:] custom_splits = [ edges_train, edges_val, edges_test, ] graph.custom_splits = custom_splits link_size_list[0] += len(edges_train) link_size_list[1] += len(edges_val) link_size_list[2] += len(edges_test) dataset = GraphDataset( graphs, task="link_pred", general_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * 2 * link_size_list[1]) self.assertEqual(split_res[2][0].edge_label_index.shape[1], 2 * 2 * link_size_list[2]) # inductive split with graph task pyg_dataset = TUDataset("./enzymes", "ENZYMES") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) num_graphs = len(graphs) split_ratio = [0.3, 0.3, 0.4] graph_size_list = [] split_offset = 0 custom_split_graphs = [] for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = (1 + int(split_ratio_i * (num_graphs - len(split_ratio)))) custom_split_graphs.append(graphs[split_offset:split_offset + num_split_i]) split_offset += num_split_i graph_size_list.append(num_split_i) else: custom_split_graphs.append(graphs[split_offset:]) graph_size_list.append(len(graphs[split_offset:])) dataset = GraphDataset( graphs, task="graph", general_split_mode="custom", custom_split_graphs=custom_split_graphs, ) split_res = dataset.split(transductive=False) self.assertEqual(graph_size_list[0], len(split_res[0])) self.assertEqual(graph_size_list[1], len(split_res[1])) self.assertEqual(graph_size_list[2], len(split_res[2])) # transductive split with link_pred task in `disjoint` edge_train_mode. pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] link_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: split_offset = 0 edges = list(graph.G.edges) random.shuffle(edges) num_edges_train = 1 + int(split_ratio[0] * (graph.num_edges - 3)) num_edges_train_disjoint = (1 + int(split_ratio[0] * 0.5 * (graph.num_edges - 3))) num_edges_val = 1 + int(split_ratio[0] * (graph.num_edges - 3)) edges_train = edges[:num_edges_train] edges_train_disjoint = edges[:num_edges_train_disjoint] edges_val = edges[num_edges_train:num_edges_train + num_edges_val] edges_test = edges[num_edges_train + num_edges_val:] custom_splits = [ edges_train, edges_val, edges_test, ] graph.custom_splits = custom_splits graph.custom_disjoint_split = edges_train_disjoint link_size_list[0] += len(edges_train_disjoint) link_size_list[1] += len(edges_val) link_size_list[2] += len(edges_test) dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", general_split_mode="custom", disjoint_split_mode="custom", ) split_res = dataset.split(transductive=True) self.assertEqual(split_res[0][0].edge_label_index.shape[1], 2 * 2 * link_size_list[0]) self.assertEqual(split_res[1][0].edge_label_index.shape[1], 2 * 2 * link_size_list[1]) self.assertEqual(split_res[2][0].edge_label_index.shape[1], 2 * 2 * link_size_list[2])
def test_dataset_split_custom(self): # transductive split with node task (self defined dataset) G, x, y, edge_x, edge_y, edge_index, graph_x, graph_y = ( simple_networkx_graph()) Graph.add_edge_attr(G, "edge_feature", edge_x) Graph.add_edge_attr(G, "edge_label", edge_y) Graph.add_node_attr(G, "node_feature", x) Graph.add_node_attr(G, "node_label", y) Graph.add_graph_attr(G, "graph_feature", graph_x) Graph.add_graph_attr(G, "graph_label", graph_y) num_nodes = len(list(G.nodes)) nodes_train = torch.tensor(list(G.nodes)[:int(0.3 * num_nodes)]) nodes_val = torch.tensor( list(G.nodes)[int(0.3 * num_nodes):int(0.6 * num_nodes)]) nodes_test = torch.tensor(list(G.nodes)[int(0.6 * num_nodes):]) graph_train = Graph(node_feature=x, node_label=y, edge_index=edge_index, node_label_index=nodes_train, directed=True) graph_val = Graph(node_feature=x, node_label=y, edge_index=edge_index, node_label_index=nodes_val, directed=True) graph_test = Graph(node_feature=x, node_label=y, edge_index=edge_index, node_label_index=nodes_test, directed=True) graphs_train = [graph_train] graphs_val = [graph_val] graphs_test = [graph_test] dataset_train, dataset_val, dataset_test = (GraphDataset(graphs_train, task='node'), GraphDataset(graphs_val, task='node'), GraphDataset(graphs_test, task='node')) self.assertEqual(dataset_train[0].node_label_index.tolist(), list(range(int(0.3 * num_nodes)))) self.assertEqual( dataset_val[0].node_label_index.tolist(), list(range(int(0.3 * num_nodes), int(0.6 * num_nodes)))) self.assertEqual(dataset_test[0].node_label_index.tolist(), list(range(int(0.6 * num_nodes), num_nodes))) # transductive split with link_pred task (train/val split) edges = list(G.edges) num_edges = len(edges) edges_train = edges[:int(0.7 * num_edges)] edges_val = edges[int(0.7 * num_edges):] link_size_list = [len(edges_train), len(edges_val)] # generate pseudo pos and neg edges, they may overlap here train_pos = torch.LongTensor(edges_train).permute(1, 0) val_pos = torch.LongTensor(edges_val).permute(1, 0) val_neg = torch.randint(high=10, size=val_pos.shape, dtype=torch.int64) val_neg_double = torch.cat((val_neg, val_neg), dim=1) num_train = len(edges_train) num_val = len(edges_val) graph_train = Graph(node_feature=x, edge_index=edge_index, edge_feature=edge_x, directed=True, edge_label_index=train_pos) graph_val = Graph(node_feature=x, edge_index=edge_index, edge_feature=edge_x, directed=True, edge_label_index=val_pos, negative_edge=val_neg_double) graphs_train = [graph_train] graphs_val = [graph_val] dataset_train, dataset_val = (GraphDataset(graphs_train, task='link_pred', resample_negatives=True), GraphDataset( graphs_val, task='link_pred', edge_negative_sampling_ratio=2)) self.assertEqual(dataset_train[0].edge_label_index.shape[1], 2 * link_size_list[0]) self.assertEqual(dataset_train[0].edge_label.shape[0], 2 * link_size_list[0]) self.assertEqual(dataset_val[0].edge_label_index.shape[1], val_pos.shape[1] + val_neg_double.shape[1]) self.assertEqual(dataset_val[0].edge_label.shape[0], val_pos.shape[1] + val_neg_double.shape[1]) self.assertTrue( torch.equal(dataset_train[0].edge_label_index[:, :num_train], train_pos)) self.assertTrue( torch.equal(dataset_val[0].edge_label_index[:, :num_val], val_pos)) self.assertTrue( torch.equal(dataset_val[0].edge_label_index[:, num_val:], val_neg_double)) dataset_train.resample_negatives = False self.assertTrue( torch.equal(dataset_train[0].edge_label_index, dataset_train[0].edge_label_index)) # transductive split with link_pred task with edge label edge_label_train = torch.LongTensor([1, 2, 3, 2, 1, 1, 2, 3, 2, 0, 0]) edge_label_val = torch.LongTensor([1, 2, 3, 2, 1, 0]) graph_train = Graph(node_feature=x, edge_index=edge_index, directed=True, edge_label_index=train_pos, edge_label=edge_label_train) graph_val = Graph(node_feature=x, edge_index=edge_index, directed=True, edge_label_index=val_pos, negative_edge=val_neg, edge_label=edge_label_val) graphs_train = [graph_train] graphs_val = [graph_val] dataset_train, dataset_val = (GraphDataset(graphs_train, task='link_pred'), GraphDataset(graphs_val, task='link_pred')) self.assertTrue( torch.equal(dataset_train[0].edge_label_index, dataset_train[0].edge_label_index)) self.assertTrue( torch.equal(dataset_train[0].edge_label[:num_train], edge_label_train)) self.assertTrue( torch.equal(dataset_val[0].edge_label[:num_val], edge_label_val)) # Multiple graph tensor backend link prediction (inductive) pyg_dataset = Planetoid('./cora', 'Cora') x = pyg_dataset[0].x y = pyg_dataset[0].y edge_index = pyg_dataset[0].edge_index row, col = edge_index mask = row < col row, col = row[mask], col[mask] edge_index = torch.stack([row, col], dim=0) edge_index = torch.cat( [edge_index, torch.flip(edge_index, [0])], dim=1) graphs = [ Graph(node_feature=x, node_label=y, edge_index=edge_index, directed=False) ] graphs = [copy.deepcopy(graphs[0]) for _ in range(10)] edge_label_index = graphs[0].edge_label_index dataset = GraphDataset(graphs, task='link_pred', edge_message_ratio=0.6, edge_train_mode="all") datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio=[0.85, 0.05, 0.1]) edge_label_index_split = ( datasets['train'][0].edge_label_index[:, 0:edge_label_index.shape[1]]) self.assertTrue(torch.equal(edge_label_index, edge_label_index_split)) # transductive split with node task (pytorch geometric dataset) pyg_dataset = Planetoid("./cora", "Cora") ds = pyg_to_dicts(pyg_dataset, task="cora") graphs = [Graph(**item) for item in ds] split_ratio = [0.3, 0.3, 0.4] node_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: custom_splits = [[] for i in range(len(split_ratio))] split_offset = 0 num_nodes = graph.num_nodes shuffled_node_indices = torch.randperm(graph.num_nodes) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = int(split_ratio_i * num_nodes) nodes_split_i = ( shuffled_node_indices[split_offset:split_offset + num_split_i]) split_offset += num_split_i else: nodes_split_i = shuffled_node_indices[split_offset:] custom_splits[i] = nodes_split_i node_size_list[i] += len(nodes_split_i) graph.custom = {"general_splits": custom_splits} node_feature = graphs[0].node_feature edge_index = graphs[0].edge_index directed = graphs[0].directed graph_train = Graph( node_feature=node_feature, edge_index=edge_index, directed=directed, node_label_index=graphs[0].custom["general_splits"][0]) graph_val = Graph( node_feature=node_feature, edge_index=edge_index, directed=directed, node_label_index=graphs[0].custom["general_splits"][1]) graph_test = Graph( node_feature=node_feature, edge_index=edge_index, directed=directed, node_label_index=graphs[0].custom["general_splits"][2]) train_dataset = GraphDataset([graph_train], task="node") val_dataset = GraphDataset([graph_val], task="node") test_dataset = GraphDataset([graph_test], task="node") self.assertEqual(len(train_dataset[0].node_label_index), node_size_list[0]) self.assertEqual(len(val_dataset[0].node_label_index), node_size_list[1]) self.assertEqual(len(test_dataset[0].node_label_index), node_size_list[2]) # transductive split with edge task pyg_dataset = Planetoid("./cora", "Cora") graphs_g = GraphDataset.pyg_to_graphs(pyg_dataset) ds = pyg_to_dicts(pyg_dataset, task="cora") graphs = [Graph(**item) for item in ds] split_ratio = [0.3, 0.3, 0.4] edge_size_list = [0 for i in range(len(split_ratio))] for i, graph in enumerate(graphs): custom_splits = [[] for i in range(len(split_ratio))] split_offset = 0 edges = list(graphs_g[i].G.edges) num_edges = graph.num_edges random.shuffle(edges) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = int(split_ratio_i * num_edges) edges_split_i = (edges[split_offset:split_offset + num_split_i]) split_offset += num_split_i else: edges_split_i = edges[split_offset:] custom_splits[i] = edges_split_i edge_size_list[i] += len(edges_split_i) graph.custom = {"general_splits": custom_splits} node_feature = graphs[0].node_feature edge_index = graphs[0].edge_index directed = graphs[0].directed train_index = torch.tensor( graphs[0].custom["general_splits"][0]).permute(1, 0) train_index = torch.cat((train_index, train_index), dim=1) val_index = torch.tensor( graphs[0].custom["general_splits"][1]).permute(1, 0) val_index = torch.cat((val_index, val_index), dim=1) test_index = torch.tensor( graphs[0].custom["general_splits"][2]).permute(1, 0) test_index = torch.cat((test_index, test_index), dim=1) graph_train = Graph(node_feature=node_feature, edge_index=edge_index, directed=directed, edge_label_index=train_index) graph_val = Graph(node_feature=node_feature, edge_index=edge_index, directed=directed, edge_label_index=val_index) graph_test = Graph(node_feature=node_feature, edge_index=edge_index, directed=directed, edge_label_index=test_index) train_dataset = GraphDataset([graph_train], task="edge") val_dataset = GraphDataset([graph_val], task="edge") test_dataset = GraphDataset([graph_test], task="edge") self.assertEqual(train_dataset[0].edge_label_index.shape[1], 2 * edge_size_list[0]) self.assertEqual(val_dataset[0].edge_label_index.shape[1], 2 * edge_size_list[1]) self.assertEqual(test_dataset[0].edge_label_index.shape[1], 2 * edge_size_list[2]) # inductive split with graph task pyg_dataset = TUDataset("./enzymes", "ENZYMES") ds = pyg_to_dicts(pyg_dataset) graphs = [Graph(**item) for item in ds] num_graphs = len(graphs) split_ratio = [0.3, 0.3, 0.4] graph_size_list = [] split_offset = 0 custom_split_graphs = [] for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = int(split_ratio_i * num_graphs) custom_split_graphs.append(graphs[split_offset:split_offset + num_split_i]) split_offset += num_split_i graph_size_list.append(num_split_i) else: custom_split_graphs.append(graphs[split_offset:]) graph_size_list.append(len(graphs[split_offset:])) dataset = GraphDataset(graphs, task="graph", custom_split_graphs=custom_split_graphs) split_res = dataset.split(transductive=False) self.assertEqual(graph_size_list[0], len(split_res[0])) self.assertEqual(graph_size_list[1], len(split_res[1])) self.assertEqual(graph_size_list[2], len(split_res[2]))
def test_dataset_split_custom(self): # transductive split with node task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] split_graphs = [[] for i in range(len(split_ratio))] node_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: split_offset = 0 shuffled_node_indices = torch.randperm(graph.num_nodes) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = ( 1 + int( split_ratio_i * (graph.num_nodes - len(split_ratio)) ) ) nodes_split_i = ( shuffled_node_indices[split_offset: split_offset + num_split_i] ) split_offset += num_split_i else: nodes_split_i = shuffled_node_indices[split_offset:] graph_new = copy.copy(graph) graph_new.custom_split_index = nodes_split_i split_graphs[i].append(graph_new) node_size_list[i] += len(nodes_split_i) dataset = GraphDataset( graphs, task="node", general_split_mode="custom", split_graphs=split_graphs ) split_res = dataset.split(transductive=True) self.assertEqual( len(split_res[0][0].node_label_index), node_size_list[0] ) self.assertEqual( len(split_res[1][0].node_label_index), node_size_list[1] ) self.assertEqual( len(split_res[2][0].node_label_index), node_size_list[2] ) # transductive split with edge task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] split_graphs = [[] for i in range(len(split_ratio))] edge_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: split_offset = 0 edges = list(graph.G.edges()) random.shuffle(edges) for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = ( 1 + int( split_ratio_i * (graph.num_edges - len(split_ratio)) ) ) edges_split_i = ( edges[split_offset: split_offset + num_split_i] ) split_offset += num_split_i else: edges_split_i = edges[split_offset:] graph_new = copy.copy(graph) graph_new.custom_split_index = edges_split_i split_graphs[i].append(graph_new) edge_size_list[i] += len(edges_split_i) dataset = GraphDataset( graphs, task="edge", general_split_mode="custom", split_graphs=split_graphs ) split_res = dataset.split(transductive=True) self.assertEqual( split_res[0][0].edge_label_index.shape[1], 2 * edge_size_list[0] ) self.assertEqual( split_res[1][0].edge_label_index.shape[1], 2 * edge_size_list[1] ) self.assertEqual( split_res[2][0].edge_label_index.shape[1], 2 * edge_size_list[2] ) # transductive split with link_pred task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) split_ratio = [0.3, 0.3, 0.4] split_graphs = [[] for i in range(len(split_ratio))] link_size_list = [0 for i in range(len(split_ratio))] for graph in graphs: split_offset = 0 edges = list(graph.G.edges(data=True)) random.shuffle(edges) num_edges_train = 1 + int(split_ratio[0] * (graph.num_edges - 3)) num_edges_val = 1 + int(split_ratio[0] * (graph.num_edges - 3)) edges_train = edges[:num_edges_train] edges_val = edges[num_edges_train:num_edges_train + num_edges_val] edges_test = edges[num_edges_train + num_edges_val:] graph_train = copy.copy(graph) graph_test = copy.copy(graph) graph_val = copy.copy(graph) graph_train.custom_split_index = edges_train graph_val.custom_split_index = edges_val graph_test.custom_split_index = edges_test split_graphs[0].append(graph_train) split_graphs[1].append(graph_val) split_graphs[2].append(graph_test) link_size_list[0] += len(edges_train) link_size_list[1] += len(edges_val) link_size_list[2] += len(edges_test) dataset = GraphDataset( graphs, task="link_pred", general_split_mode="custom", split_graphs=split_graphs ) split_res = dataset.split(transductive=True) self.assertEqual( split_res[0][0].edge_label_index.shape[1], 2 * 2 * link_size_list[0] ) self.assertEqual( split_res[1][0].edge_label_index.shape[1], 2 * 2 * link_size_list[1] ) self.assertEqual( split_res[2][0].edge_label_index.shape[1], 2 * 2 * link_size_list[2] ) # inductive split with graph task pyg_dataset = TUDataset("./enzymes", "ENZYMES") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) num_graphs = len(graphs) split_ratio = [0.3, 0.3, 0.4] graph_size_list = [] split_offset = 0 split_graphs = [] for i, split_ratio_i in enumerate(split_ratio): if i != len(split_ratio) - 1: num_split_i = ( 1 + int(split_ratio_i * (num_graphs - len(split_ratio))) ) split_graphs.append( graphs[split_offset: split_offset + num_split_i] ) split_offset += num_split_i graph_size_list.append(num_split_i) else: split_graphs.append(graphs[split_offset:]) graph_size_list.append(len(graphs[split_offset:])) dataset = GraphDataset( graphs, task="graph", general_split_mode="custom", split_graphs=split_graphs ) split_res = dataset.split(transductive=False) self.assertEqual(graph_size_list[0], len(split_res[0])) self.assertEqual(graph_size_list[1], len(split_res[1])) self.assertEqual(graph_size_list[2], len(split_res[2]))
def test_dataset_split(self): # inductively split with graph task pyg_dataset = TUDataset("./enzymes", "ENZYMES") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="graph") split_res = dataset.split(transductive=False) num_graphs = len(dataset) num_graphs_reduced = num_graphs - 3 num_train = 1 + int(num_graphs_reduced * 0.8) num_val = 1 + int(num_graphs_reduced * 0.1) num_test = num_graphs - num_train - num_val self.assertEqual(num_train, len(split_res[0])) self.assertEqual(num_val, len(split_res[1])) self.assertEqual(num_test, len(split_res[2])) # inductively split with link_pred task # and default (`all`) edge_train_mode pyg_dataset = TUDataset("./enzymes", "ENZYMES") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="link_pred") split_res = dataset.split(transductive=False) num_graphs = len(dataset) num_graphs_reduced = num_graphs - 3 num_train = 1 + int(num_graphs_reduced * 0.8) num_val = 1 + int(num_graphs_reduced * 0.1) num_test = num_graphs - num_train - num_val self.assertEqual(num_train, len(split_res[0])) self.assertEqual(num_val, len(split_res[1])) self.assertEqual(num_test, len(split_res[2])) # inductively split with link_pred task and `disjoint` edge_train_mode pyg_dataset = TUDataset("./enzymes", "ENZYMES") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", ) split_res = dataset.split(transductive=False) num_graphs = len(dataset) num_graphs_reduced = num_graphs - 3 num_train = 1 + int(num_graphs_reduced * 0.8) num_val = 1 + int(num_graphs_reduced * 0.1) num_test = num_graphs - num_train - num_val self.assertEqual(num_train, len(split_res[0])) self.assertEqual(num_val, len(split_res[1])) self.assertEqual(num_test, len(split_res[2])) # transductively split with node task pyg_dataset = Planetoid("./cora", "Cora") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="node") num_nodes = dataset.num_nodes[0] num_nodes_reduced = num_nodes - 3 num_edges = dataset.num_edges[0] num_edges_reduced = num_edges - 3 split_res = dataset.split() self.assertEqual( len(split_res[0][0].node_label_index), 1 + int(0.8 * num_nodes_reduced) ) self.assertEqual( len(split_res[1][0].node_label_index), 1 + int(0.1 * num_nodes_reduced) ) self.assertEqual( len(split_res[2][0].node_label_index), num_nodes - 2 - int(0.8 * num_nodes_reduced) - int(0.1 * num_nodes_reduced) ) # transductively split with edge task dataset = GraphDataset(graphs, task="edge") split_res = dataset.split() edge_0 = 2 * (1 + int(0.8 * (num_edges_reduced))) self.assertEqual( split_res[0][0].edge_label_index.shape[1], edge_0, ) edge_1 = 2 * (1 + int(0.1 * (num_edges_reduced))) self.assertEqual( split_res[1][0].edge_label_index.shape[1], edge_1, ) self.assertEqual( split_res[2][0].edge_label_index.shape[1], 2 * num_edges - edge_0 - edge_1, ) # transductively split with link_pred task # and default (`all`) edge_train_mode dataset = GraphDataset(graphs, task="link_pred") split_res = dataset.split() self.assertEqual( split_res[0][0].edge_label_index.shape[1], 2 * 2 * (1 + int(0.8 * (num_edges_reduced))) ) self.assertEqual( split_res[1][0].edge_label_index.shape[1], 2 * 2 * (1 + (int(0.1 * (num_edges_reduced)))) ) self.assertEqual( split_res[2][0].edge_label_index.shape[1], 2 * 2 * num_edges - 2 * 2 * ( 2 + int(0.1 * num_edges_reduced) + int(0.8 * num_edges_reduced) ) ) # transductively split with link_pred task, `split` edge_train_mode # and 0.5 edge_message_ratio dataset = GraphDataset( graphs, task="link_pred", edge_train_mode="disjoint", edge_message_ratio=0.5, ) split_res = dataset.split() edge_0 = 2 * (1 + int(0.8 * num_edges_reduced)) edge_0 = 2 * (edge_0 - (1 + int(0.5 * (edge_0 - 3)))) self.assertEqual( split_res[0][0].edge_label_index.shape[1], edge_0, ) edge_1 = 2 * 2 * (1 + int(0.1 * num_edges_reduced)) self.assertEqual(split_res[1][0].edge_label_index.shape[1], edge_1) edge_2 = ( 2 * 2 * int(num_edges) - 2 * 2 * (1 + int(0.8 * num_edges_reduced)) - edge_1 ) self.assertEqual(split_res[2][0].edge_label_index.shape[1], edge_2) # transductively split with link_pred task # and specified edge_negative_sampling_ratio dataset = GraphDataset( graphs, task="link_pred", edge_negative_sampling_ratio=2 ) split_res = dataset.split() edge_0 = (2 + 1) * (2 * (1 + int(0.8 * num_edges_reduced))) self.assertEqual(split_res[0][0].edge_label_index.shape[1], edge_0) edge_1 = (2 + 1) * 2 * (1 + int(0.1 * num_edges_reduced)) self.assertEqual(split_res[1][0].edge_label_index.shape[1], edge_1) edge_2 = (2 + 1) * 2 * int(num_edges) - edge_0 - edge_1 self.assertEqual(split_res[2][0].edge_label_index.shape[1], edge_2)
acc /= total return acc if __name__ == "__main__": args = arg_parse() if args.dataset in ['Cora', 'CiteSeer', 'Pubmed']: pyg_dataset = Planetoid( './planetoid', args.dataset, transform=T.TargetIndegree()) # load some format of graph data else: raise ValueError("Unsupported dataset.") if args.split == 'random': graphs = GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=False, tensor_backend=True) dataset = GraphDataset(graphs, task='node') # node, edge, link_pred, graph dataset_train, dataset_val, dataset_test = dataset.split( transductive=True, split_ratio=[0.8, 0.1, 0.1]) # transductive split, inductive split else: graphs_train, graphs_val, graphs_test = \ GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=True, tensor_backend=True) dataset_train, dataset_val, dataset_test = \ GraphDataset(graphs_train, task='node'), GraphDataset(graphs_val,task='node'), \ GraphDataset(graphs_test, task='node')
import pdb from deepsnap.dataset import GraphDataset from deepsnap.batch import Batch from torch.utils.data import DataLoader name = 'Cora' model_name = 'GCN' fixed_split = True pyg_dataset = Planetoid( './cora', name, transform=T.TargetIndegree()) # load some format of graph data if not fixed_split: graphs = GraphDataset.pyg_to_graphs( pyg_dataset, verbose=True, fixed_split=fixed_split) # transform to our format dataset = GraphDataset(graphs, task='node') # node, edge, link_pred, graph dataset_train, dataset_val, dataset_test = dataset.split( transductive=True, split_ratio=[0.8, 0.1, 0.1]) # transductive split, inductive split else: graphs_train, graphs_val, graphs_test = \ GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=fixed_split) # transform to our format dataset_train, dataset_val, dataset_test = \ GraphDataset(graphs_train, task='node'), GraphDataset(graphs_val, task='node'), GraphDataset(graphs_test, task='node') train_loader = DataLoader(dataset_train,
num_graphs += batch.num_graphs # print("loader len {}".format(num_graphs)) return correct / num_graphs if __name__ == "__main__": args = arg_parse() if args.dataset == 'enzymes': pyg_dataset = TUDataset('./enzymes', 'ENZYMES') elif args.dataset == 'dd': pyg_dataset = TUDataset('./dd', 'DD') else: raise ValueError("Unsupported dataset.") graphs = GraphDataset.pyg_to_graphs(pyg_dataset) dataset = GraphDataset(graphs, task="graph") datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio=[0.8, 0.1, 0.1]) dataloaders = { split: DataLoader(dataset, collate_fn=Batch.collate(), batch_size=args.batch_size, shuffle=True) for split, dataset in datasets.items() } num_classes = datasets['train'].num_graph_labels num_node_features = datasets['train'].num_node_features
acc = pred.eq(batch.node_label[batch.node_label_index]).sum().item() total = batch.node_label_index.shape[0] acc /= total return acc if __name__ == "__main__": args = arg_parse() if args.dataset in ['Cora', 'CiteSeer', 'Pubmed']: pyg_dataset = Planetoid('./planetoid', args.dataset, transform=T.TargetIndegree()) # load some format of graph data else: raise ValueError("Unsupported dataset.") if args.split == 'random': graphs = GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=False, netlib=netlib) # transform to our format dataset = GraphDataset(graphs, task='node') # node, edge, link_pred, graph dataset_train, dataset_val, dataset_test = dataset.split( transductive=True, split_ratio=[0.8, 0.1, 0.1]) # transductive split, inductive split else: graphs_train, graphs_val, graphs_test = \ GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=True, netlib=netlib) # transform to our format dataset_train, dataset_val, dataset_test = \ GraphDataset(graphs_train, task='node'), GraphDataset(graphs_val,task='node'), \ GraphDataset(graphs_test, task='node')
label = graph.node_label pred = pred[graph.node_label_index].max(1)[1] acc = pred.eq(label).sum().item() acc /= len(label) accs.append(acc) return accs if __name__ == '__main__': args = arg_parse() pyg_dataset = Planetoid('./planetoid', 'Cora') # Full batch print("Full batch training") graphs_train, graphs_val, graphs_test = \ GraphDataset.pyg_to_graphs(pyg_dataset, verbose=True, fixed_split=True) graph_train = graphs_train[0] graph_val = graphs_val[0] graph_test = graphs_test[0] model = GNN(graph_train.num_node_features, args.hidden_size, graph_train.num_node_labels, args).to(args.device) optimizer = torch.optim.Adam(model.parameters(), lr=args.lr) graphs = [graph_train, graph_val, graph_test] all_best_model, all_accs = train(graphs, graphs, args, model, optimizer, mode="all") train_acc, val_acc, test_acc = test([graph_train, graph_val, graph_test], all_best_model)
label = batch.graph_label correct += pred.eq(label).sum().item() num_graphs += batch.num_graphs return correct / num_graphs if __name__ == "__main__": args = arg_parse() if args.dataset == 'enzymes': pyg_dataset = TUDataset('./enzymes', 'ENZYMES') elif args.dataset == 'dd': pyg_dataset = TUDataset('./dd', 'DD') else: raise ValueError("Unsupported dataset.") graphs = GraphDataset.pyg_to_graphs(pyg_dataset, tensor_backend=True) dataset = GraphDataset(graphs, task="graph") datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio = [0.8, 0.1, 0.1]) dataloaders = {split: DataLoader( dataset, collate_fn=Batch.collate(), batch_size=args.batch_size, shuffle=True) for split, dataset in datasets.items()} num_classes = datasets['train'].num_graph_labels num_node_features = datasets['train'].num_node_features train(dataloaders['train'], dataloaders['val'], dataloaders['test'], args, num_node_features, num_classes, args.device)
else: raise ValueError("Unsupported dataset.") if args.netlib == "nx": import networkx as netlib print("Use NetworkX as the backend network library.") elif args.netlib == "sx": import snap import snapx as netlib print("Use SnapX as the backend network library.") else: raise ValueError("{} network library is not supported.".format(args.netlib)) args.netlib = netlib graphs = GraphDataset.pyg_to_graphs(pyg_dataset, netlib=args.netlib) dataset = GraphDataset(graphs, task="graph") datasets = {} datasets['train'], datasets['val'], datasets['test'] = dataset.split( transductive=False, split_ratio = [0.8, 0.1, 0.1]) if args.transform_dataset is not None: trans_func = get_transform(args.transform_dataset) for _, dataset in datasets.items(): dataset.apply_transform(trans_func, radius=args.radius, netlib=args.netlib) dataloaders = { split: DataLoader( dataset, collate_fn=Batch.collate(), batch_size=args.batch_size, shuffle=True