def train(step, model, optimizer, lr_scheduler, mm_scheduler, transform_1,
transform_2, data, args):
model.train()
# update learning rate
lr = lr_scheduler.get(step)
for param_group in optimizer.param_groups:
param_group['lr'] = lr
# update momentum
mm = 1 - mm_scheduler.get(step)
# forward
optimizer.zero_grad()
x1, x2 = transform_1(data), transform_2(data)
if args.dataset != 'ppi':
x1, x2 = dgl.add_self_loop(x1), dgl.add_self_loop(x2)
q1, y2 = model(x1, x2)
q2, y1 = model(x2, x1)
loss = 2 - cosine_similarity(q1, y2.detach(),
dim=-1).mean() - cosine_similarity(
q2, y1.detach(), dim=-1).mean()
loss.backward()
# update online network
optimizer.step()
# update target network
model.update_target_network(mm)
return loss.item()
def data_split(g, args):
index = data_sample(g, args)
# num_nodes = g.number_of_nodes()
# train_mask = g.ndata['train_mask']
# val_mask = g.ndata['val_mask']
# test_mask = g.ndata['test_mask']
# train_ind = torch.where(train_mask == True)[0].tolist()
# val_ind = torch.where(val_mask == True)[0].tolist()
# test_ind = torch.where(test_mask == True)[0].tolist()
# else_ind = list(set(range(num_nodes)) -
# set(train_ind) - set(val_ind) - set(test_ind))
# shuffle(train_ind, val_ind, test_ind, else_ind)
# shard = [int(len(train_ind)/args.split), int(len(val_ind)/args.split),
# int(len(test_ind)/args.split), int(len(else_ind)/args.split)]
# ind = [train_ind[i*shard[0]:(i+1)*shard[0]] + val_ind[i*shard[1]:(i+1)*shard[1]] + test_ind[i*shard[2]:(
# i+1)*shard[2]] + else_ind[i*shard[3]:(i+1)*shard[3]] for i in range(args.split)]
graphs = [node_subgraph(g, index[i]) for i in range(args.split)]
for i in range(len(graphs)):
# graphs[i].int().to(args.device)
# add self loop
graphs[i] = remove_self_loop(graphs[i])
graphs[i] = add_self_loop(graphs[i])
return graphs
def forward(self, g, feats):
"""Update node representations.
Parameters
----------
g : DGLGraph
DGLGraph for a batch of graphs
feats : FloatTensor of shape (N, M1)
* N is the total number of nodes in the batch of graphs
* M1 is the input node feature size, which equals in_feats in initialization
Returns
-------
feats : FloatTensor of shape (N, M2)
* N is the total number of nodes in the batch of graphs
* M2 is the output node representation size, which equals
hidden_sizes[-1] in initialization.
"""
deg = None
max_deg = None
deg_membership = None
g_self = dgl.add_self_loop(g)
for gnn in self.gnn_layers:
feats, deg, max_deg, deg_membership = gnn(g, g_self, feats, deg,
max_deg, deg_membership)
return feats
def DGLDatasetReader(dataset_name, self_loops, device=None):
data = load_data(dataset_name, self_loops)
if dataset_name == 'reddit':
g = data.graph.int().to(device)
# normalization
degs = g.in_degrees().float()
norm = torch.pow(degs, -0.5)
norm[torch.isinf(norm)] = 0
g.ndata['norm'] = norm.unsqueeze(1)
else:
g = DGLGraph(data.graph).to(device)
# normalization
degs = g.in_degrees().float()
norm = torch.pow(degs, -0.5)
norm[torch.isinf(norm)] = 0
norm = norm.to(device)
g.ndata['norm'] = norm.unsqueeze(1)
# add self loop
if self_loops:
# g.add_edges(g.nodes(), g.nodes())
g = dgl.remove_self_loop(g)
g = dgl.add_self_loop(g)
return g,torch.FloatTensor(data.features),torch.LongTensor(data.labels),data.num_labels,\
torch.ByteTensor(data.train_mask),torch.ByteTensor(data.test_mask),torch.ByteTensor(data.val_mask)
def train(args):
data = load_citation_dataset(args.dataset)
g = data[0]
features = g.ndata['feat']
labels = g.ndata['label']
train_mask = g.ndata['train_mask']
val_mask = g.ndata['val_mask']
test_mask = g.ndata['test_mask']
# add self loop
g = dgl.remove_self_loop(g)
g = dgl.add_self_loop(g)
num_heads = [args.num_heads] * (args.num_layers - 1) + [args.num_out_heads]
model = GAT(features.shape[1], args.num_hidden, data.num_classes,
num_heads, args.dropout)
optimizer = optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
for epoch in range(args.epochs):
model.train()
logits = model(g, features)
loss = F.cross_entropy(logits[train_mask], labels[train_mask])
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_acc = accuracy(logits[train_mask], labels[train_mask])
val_acc = evaluate(model, g, features, labels, val_mask)
print('Epoch {:04d} | Loss {:.4f} | Train Acc {:.4f} | Val Acc {:.4f}'.
format(epoch, loss.item(), train_acc, val_acc))
print()
acc = evaluate(model, g, features, labels, test_mask)
print('Test Accuracy {:.4f}'.format(acc))
def train(args):
set_random_seed(args.seed)
device = get_device(args.device)
data, g, feats, labels, predict_ntype, relations, neighbor_sizes, \
pos, pos_threshold, train_mask, val_mask, test_mask = load_data(args.dataset, device)
bgs = [g[rel] for rel in relations] # 邻居-目标顶点二分图
mgs = [
dgl.add_self_loop(
dgl.remove_self_loop(dgl.metapath_reachable_graph(g,
mp))).to(device)
for mp in data.metapaths
] # 基于元路径的邻居同构图
model = HeCo([feat.shape[1] for feat in feats], args.num_hidden,
args.feat_drop, args.attn_drop, neighbor_sizes,
len(data.metapaths), args.tau, args.lambda_).to(device)
optimizer = optim.Adam(model.parameters(), lr=args.lr)
for epoch in range(args.epochs):
model.train()
loss = model(bgs, mgs, feats, pos)
optimizer.zero_grad()
loss.backward()
optimizer.step()
print('Epoch {:d} | Train Loss {:.4f}'.format(epoch, loss.item()))
evaluate(model, mgs, feats[0], labels, data.num_classes, train_mask,
test_mask, args.seed)
def add_recipe(step):
g, mp = step
n = g.number_of_nodes()
items = list(range(18, n, 1))
size = np.random.randint(len(items))
us = np.random.choice(items, size=size)
us = torch.tensor(us)
us = torch.tensor(us, dtype=torch.int32).to(device)
vs = np.zeros((size, )) + (n - 1)
vs = torch.tensor(vs, dtype=torch.int32).to(device)
norms = torch.ones(len(vs))
norms = torch.tensor(norms, dtype=torch.float32).to(device)
types = torch.tensor(torch.ones(len(vs)).to(device) + 3,
dtype=torch.int32).to(device)
g.add_edges(us, vs, data={"rel_type": types, "norm": norms})
g.add_edges(vs, us, data={"rel_type": types, "norm": norms})
g = dgl.add_self_loop(g)
return (g, mp)
def dgl_graph_from_vec(vec, graph_params):
"""
Create graph from flatten vector as a thresholed weighted matrix with properties
as type torch
"""
if graph_params.flatten:
W = vec_to_sym(vec)
else:
W = vec
# create graph
# add signal on nodes
u = getattr(feature_generation, graph_params.node_feat)(W)
if graph_params.thr_type == 'pos':
W[W < graph_params.threshold] = 0
else:
W[np.abs(W) < graph_params.threshold] = 0
# convert to pytorch?
W = sparse.csr_matrix(W).tocoo()
edge_weight = torch.tensor(W.data).float()
u = torch.from_numpy(u.astype(np.float32))
g = dgl.from_scipy(W)
g.ndata['feat'] = u
g.edata['weight'] = edge_weight
if graph_params.add_self_loop:
g = dgl.add_self_loop(g)
g.edata['weight'][-graph_params.n_nodes:] = 1
return g
def __call__(self, split_type):
if split_type == 'train':
subsample_ratio = self.subsample_ratio
else:
subsample_ratio = 1
pos_edges = self.split_edge[split_type]['edge']
if split_type == 'train':
# Adding self loop in train avoids sampling the source node itself.
g = add_self_loop(self.g)
eids = g.edge_ids(pos_edges[:, 0], pos_edges[:, 1])
neg_edges = torch.stack(self.neg_sampler(g, eids), dim=1)
else:
neg_edges = self.split_edge[split_type]['edge_neg']
pos_edges = self.subsample(pos_edges, subsample_ratio).long()
neg_edges = self.subsample(neg_edges, subsample_ratio).long()
edges = torch.cat([pos_edges, neg_edges])
labels = torch.cat([
torch.ones(pos_edges.size(0), 1),
torch.zeros(neg_edges.size(0), 1)
])
if self.shuffle:
perm = torch.randperm(edges.size(0))
edges = edges[perm]
labels = labels[perm]
return edges, labels
def gat_data(g, n_classes, gpu):
# add self loop
g = dgl.remove_self_loop(g)
g = dgl.add_self_loop(g)
n_edges = g.number_of_edges()
num_feats = g.ndata['feat'].shape[1]
return n_classes, n_edges, g, num_feats
def to_dgl(self: GraphFeaturiser, mol: Mol) -> dgl.DGLGraph:
"""Generates a DGL graph from a molecule.
Args:
mol: The molecule to featurise.
Returns:
A DGL graph of the featurised molecule.
"""
num_atoms = mol.GetNumAtoms()
bonds = mol.GetBonds()
bond_from = [bond.GetBeginAtomIdx() for bond in bonds]
bond_to = [bond.GetEndAtomIdx() for bond in bonds]
g = dgl.graph((torch.tensor(bond_from), torch.tensor(bond_to)),
num_nodes=num_atoms)
for key, atom_featuriser in self.atom_featurisers.items():
atom_features = atom_featuriser.process_molecule(mol)
g.ndata[key] = torch.tensor(atom_features, dtype=torch.float)
for key, bond_featuriser in self.bond_featurisers.items():
bond_features = [
bond_featuriser.process_bond(bond) for bond in bonds
]
g.edata[key] = torch.tensor(bond_features, dtype=torch.float)
g = dgl.add_reverse_edges(g, copy_edata=True)
if self.add_self_loops:
g = dgl.add_self_loop(g)
return g
def get_graph(glass_tc_pos_path,
control_tc_pos_path,
threshold,
weight,
only_control: bool = True):
adj_matrix, pos_glass, pos_control = get_adj_matrix(
glass_tc_pos_path, control_tc_pos_path, threshold, weight,
only_control)
num_state_node = pos_glass.shape[0]
total_node = adj_matrix.shape[0]
u = torch.nonzero(adj_matrix)[:, 0]
v = torch.nonzero(adj_matrix)[:, 1]
# 0th node ~ 'state dim-1'th node : state node (glass TC)
# Others : action node (control TC)
g = dgl.graph((u, v), num_nodes=total_node)
g = dgl.add_self_loop(g)
# Add control node flags
is_control = torch.zeros(total_node, 1)
is_control[num_state_node:, 0] = 1
g.ndata['is_control'] = is_control
# handling positional information.
scaler = MinMaxScaler()
pos = np.concatenate([pos_glass, pos_control], axis=0)
pos_std = scaler.fit_transform(pos)
g.ndata['position'] = torch.from_numpy(pos_std).float()
return g
def detect_body_point_and_save(img_filenames, imgs, save_name, label):
# 若已存在对应的.npy，则直接导入，不再重新检测关键点
if os.path.exists(save_name):
print('Body Point data have already existed')
bodies_points = np.load(save_name)
else:
# 若无对应.npy，则基于openpose进行人体关键点检测
bodies_points = []
for img in imgs:
# print(img_filenames + img)
body_point = get_body_points(img_filenames + img)
bodies_points.append(body_point)
bodies_points_array = np.array(bodies_points)
np.save(save_name, bodies_points_array)
body_graph = [] # 空列表用于存储graph
for body_points in bodies_points: # 遍历所有检测图片结果
for body_point in body_points: # 由于某些图片中存在多个人，所以再次再同一张图片结果中遍历所有人体
g = dgl.DGLGraph((src, dst)) # 建立graph
g = dgl.add_self_loop(g)
plt.figure()
pos = body_point[:, 0:2] # 取前两列坐标（第三列为置信度），作为显示时的节点坐标
pos[:, 1] = -pos[:, 1] # 由于默认显示坐标原点再左上角，鉴于人眼观察习惯，将其坐标上下翻转
nx.draw(g.to_networkx(), pos=pos, with_labels=True) # 显示graph
node_feature = body_point
g.ndata['coordinate'] = torch.tensor(node_feature)
body_graph.append([g, torch.tensor([label]).float()])
return body_graph
def compute_representations(net, dataset, device):
r"""Pre-computes the representations for the entire data.
Returns:
[torch.Tensor, torch.Tensor]: Representations and labels.
"""
net.eval()
reps = []
labels = []
if len(dataset) == 1:
g = dataset[0]
g = dgl.add_self_loop(g)
g = g.to(device)
with torch.no_grad():
reps.append(net(g))
labels.append(g.ndata['label'])
else:
for g in dataset:
# forward
g = g.to(device)
with torch.no_grad():
reps.append(net(g))
labels.append(g.ndata['label'])
reps = torch.cat(reps, dim=0)
labels = torch.cat(labels, dim=0)
return [reps, labels]
def train_gcn(args):
exp_init(args.seed, gpu_id=args.gpu)
# ! config
cf = GraphSageConfig(args)
cf.device = th.device("cuda:0" if args.gpu >= 0 else "cpu")
# ! Load Graph
g, features, n_feat, cf.n_class, labels, train_x, val_x, test_x = preprocess_data(cf.dataset, cf.train_percentage)
features = features.to(cf.device)
g = dgl.add_self_loop(g).to(cf.device)
supervision = SimpleObject({'train_x': train_x, 'val_x': val_x, 'test_x': test_x, 'labels': labels})
# ! Train Init
print(f'{cf}\nStart training..')
model = SAGE(n_feat, cf.n_hidden, cf.n_class, cf.n_layer, F.relu, cf.dropout, cf.aggregator)
model.to(cf.device)
print(model)
optimizer = th.optim.Adam(
model.parameters(), lr=cf.lr, weight_decay=cf.weight_decay)
if cf.early_stop > 0:
stopper = EarlyStopping(patience=cf.early_stop, path=cf.checkpoint_file)
else:
stopper = None
# ! Train
trainer = FullBatchTrainer(model=model, g=g, cf=cf, features=features,
sup=supervision, stopper=stopper, optimizer=optimizer,
loss_func=th.nn.CrossEntropyLoss())
trainer.run()
trainer.eval_and_save()
return cf
def train(args):
data = load_citation_dataset(args.dataset)
g = data[0]
features = g.ndata['feat']
labels = g.ndata['label']
train_mask = g.ndata['train_mask']
val_mask = g.ndata['val_mask']
test_mask = g.ndata['test_mask']
# add self loop
g = dgl.remove_self_loop(g)
g = dgl.add_self_loop(g)
model = GCN(features.shape[1], args.num_hidden, data.num_classes)
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
for epoch in range(args.epochs):
model.train()
logits = model(g, features)
loss = criterion(logits[train_mask], labels[train_mask])
optimizer.zero_grad()
loss.backward()
optimizer.step()
acc = accuracy(logits[val_mask], labels[val_mask])
print('Epoch {:05d} | Loss {:.4f} | ValAcc {:.4f}'.format(
epoch, loss.item(), acc))
acc = accuracy(model(g, features)[test_mask], labels[test_mask])
print('Test Accuracy {:.4f}'.format(acc))
def train(args):
set_random_seed(args.seed)
if args.hetero:
data = HETERO_DATASET[args.dataset]()
g = data[0]
gs = [
dgl.metapath_reachable_graph(g, metapath)
for metapath in data.metapaths
]
for i in range(len(gs)):
gs[i] = dgl.add_self_loop(dgl.remove_self_loop(gs[i]))
ntype = data.predict_ntype
num_classes = data.num_classes
features = g.nodes[ntype].data['feat']
labels = g.nodes[ntype].data['label']
train_mask = g.nodes[ntype].data['train_mask']
val_mask = g.nodes[ntype].data['val_mask']
test_mask = g.nodes[ntype].data['test_mask']
else:
data = DATASET[args.dataset]()
gs = data[0]
num_classes = data.num_classes
features = gs[0].ndata['feat']
labels = gs[0].ndata['label']
train_mask = gs[0].ndata['train_mask']
val_mask = gs[0].ndata['val_mask']
test_mask = gs[0].ndata['test_mask']
model = HAN(len(gs), features.shape[1], args.num_hidden, num_classes,
args.num_heads, args.dropout)
optimizer = optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
score = micro_macro_f1_score if args.task == 'clf' else nmi_ari_score
if args.task == 'clf':
metrics = 'Epoch {:d} | Train Loss {:.4f} | Train Micro-F1 {:.4f} | Train Macro-F1 {:.4f}' \
' | Val Micro-F1 {:.4f} | Val Macro-F1 {:.4f}'
else:
metrics = 'Epoch {:d} | Train Loss {:.4f} | Train NMI {:.4f} | Train ARI {:.4f}' \
' | Val NMI {:.4f} | Val ARI {:.4f}'
for epoch in range(args.epochs):
model.train()
logits = model(gs, features)
loss = F.cross_entropy(logits[train_mask], labels[train_mask])
optimizer.zero_grad()
loss.backward()
optimizer.step()
train_metrics = score(logits[train_mask], labels[train_mask])
val_metrics = score(logits[val_mask], labels[val_mask])
print(metrics.format(epoch, loss.item(), *train_metrics, *val_metrics))
test_metrics = evaluate(model, gs, features, labels, test_mask, score)
if args.task == 'clf':
print('Test Micro-F1 {:.4f} | Test Macro-F1 {:.4f}'.format(
*test_metrics))
else:
print('Test NMI {:.4f} | Test ARI {:.4f}'.format(*test_metrics))
def create_elliptic_dataset_graph(df_features,
df_edges,
df_features_orig=None,
encode_nodes=False,
create_masks=False,
graph_type='pytorch-geometric'):
'''
df_features_orig - датафрейм фичей, который нужно использовать в качестве фичей в графе (для случая с EvolveGCN, когда требуется не меняющаяся матрица фичей в темпоральном датасете)
'''
assert graph_type in ['pytorch-geometric', 'dgl']
df_features_orig = df_features if df_features_orig is None else df_features_orig
df_edges = df_edges.copy()
df_edges = df_edges[df_edges['txId1'].isin(df_features[0])
| df_edges['txId2'].isin(df_features[0])]
edge_index = df_edges.to_numpy().T
edge_index = torch.LongTensor(edge_index).contiguous()
weights = None
node_features = df_features_orig.copy()
y = torch.LongTensor(node_features['class'].values)
if encode_nodes:
node_encoder = OrderedEncoder()
node_encoder.fit(node_features[0])
edge_index = torch.LongTensor(node_encoder.transform(edge_index))
node_features = node_features.drop([0, 'class', 1], axis=1)
node_features = torch.FloatTensor(node_features.values)
if graph_type == 'pytorch-geometric':
data = torch_geometric_graph(x=node_features,
edge_index=edge_index,
edge_attr=weights,
y=y)
num_nodes = data.num_nodes
else:
U, V = edge_index[0, :], edge_index[1, :]
data = dgl_graph((U, V))
data.ndata['feat'] = node_features
data.ndata['label'] = y
data = dgl.add_self_loop(data)
num_nodes = data.num_nodes()
if create_masks:
train_idx, test_idx = train_test_split(np.arange(num_nodes),
test_size=0.15,
random_state=42,
stratify=data.get_node_labels())
data.train_mask = torch.BoolTensor([(node in train_idx)
for node in np.arange(num_nodes)])
data.test_mask = torch.BoolTensor([(node in test_idx)
for node in np.arange(num_nodes)])
return data
def __init__(self, batch_size: int):
super().__init__()
dataset = DglNodePropPredDataset(name='ogbn-arxiv')
self.split_idx = dataset.get_idx_split()
self.g, labels = dataset[0]
self.g.ndata["label"] = labels.squeeze()
self.g = add_self_loop(self.g)
self.batch_size = batch_size
def networkx_to_torch(self, networkx_graph):
import dgl
# graph = dgl.DGLGraph()
graph = dgl.from_networkx(networkx_graph)
graph = dgl.remove_self_loop(graph)
graph = dgl.add_self_loop(graph)
graph = graph.to(self.device)
return graph
def main(args):
# Step 1: Prepare graph data and retrieve train/validation/test index ============================= #
dataset = LegacyTUDataset(args.dataset, raw_dir=args.dataset_path)
# add self loop. We add self loop for each graph here since the function "add_self_loop" does not
# support batch graph.
for i in range(len(dataset)):
dataset.graph_lists[i] = dgl.add_self_loop(dataset.graph_lists[i])
num_training = int(len(dataset) * 0.8)
num_val = int(len(dataset) * 0.1)
num_test = len(dataset) - num_val - num_training
train_set, val_set, test_set = random_split(dataset, [num_training, num_val, num_test])
train_loader = GraphDataLoader(train_set, batch_size=args.batch_size, shuffle=True, num_workers=6)
val_loader = GraphDataLoader(val_set, batch_size=args.batch_size, num_workers=2)
test_loader = GraphDataLoader(test_set, batch_size=args.batch_size, num_workers=2)
device = torch.device(args.device)
# Step 2: Create model =================================================================== #
num_feature, num_classes, _ = dataset.statistics()
model_op = get_sag_network(args.architecture)
model = model_op(in_dim=num_feature, hid_dim=args.hid_dim, out_dim=num_classes,
num_convs=args.conv_layers, pool_ratio=args.pool_ratio, dropout=args.dropout).to(device)
args.num_feature = int(num_feature)
args.num_classes = int(num_classes)
# Step 3: Create training components ===================================================== #
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.weight_decay)
# Step 4: training epoches =============================================================== #
bad_cound = 0
best_val_loss = float("inf")
final_test_acc = 0.
best_epoch = 0
train_times = []
for e in range(args.epochs):
s_time = time()
train_loss = train(model, optimizer, train_loader, device)
train_times.append(time() - s_time)
val_acc, val_loss = test(model, val_loader, device)
test_acc, _ = test(model, test_loader, device)
if best_val_loss > val_loss:
best_val_loss = val_loss
final_test_acc = test_acc
bad_cound = 0
best_epoch = e + 1
else:
bad_cound += 1
if bad_cound >= args.patience:
break
if (e + 1) % args.print_every == 0:
log_format = "Epoch {}: loss={:.4f}, val_acc={:.4f}, final_test_acc={:.4f}"
print(log_format.format(e + 1, train_loss, val_acc, final_test_acc))
print("Best Epoch {}, final test acc {:.4f}".format(best_epoch, final_test_acc))
return final_test_acc, sum(train_times) / len(train_times)
def subsampled_batched_mesh_dgl(self, batchsize, faces_percentage=0.7):
self.__subsample_faces(faces_percentage)
mesh = self.__dgl_mesh()
batch = [mesh for i in range(batchsize)]
batched_mesh = dgl.batch(batch)
batched_mesh = dgl.add_self_loop(batched_mesh)
return batched_mesh
def giveGraphs(self, batch_size, voxel_pos):
p2v = np.load("data/p2v_spec.npy", allow_pickle=True).tolist()
p2v = [item for sublist in p2v for item in sublist]
p2p = np.load("data/p2p.npy", allow_pickle=True).tolist()
p2p = [item for sublist in p2p for item in sublist]
v2v = np.load("data/v2v.npy", allow_pickle=True).tolist()
v2v = [item for sublist in v2v for item in sublist]
v2v_6 = np.load("data/v2v_6.npy", allow_pickle=True).tolist()
v2v_6 = [item for sublist in v2v_6 for item in sublist]
G_vox = dgl.graph(v2v)
G_vox = dgl.add_self_loop(G_vox)
graph_data = {('PMT', 'p2v', 'vox'): p2v, ('vox', 'v2v', 'vox'): v2v}
g = dgl.heterograph(graph_data)
g = dgl.to_homogeneous(g)
g = dgl.add_self_loop(g)
G = dgl.batch([g for i in range(batch_size)])
return G, G_vox
def load_dataset():
dataset = DglNodePropPredDataset(name='ogbn-arxiv')
split_idx = dataset.get_idx_split()
# there is only one graph in Node Property Prediction datasets
g, labels = dataset[0]
g = dgl.add_self_loop(g)
return g, labels, split_idx
def main(args):
# Step 1: Prepare graph data and retrieve train/validation/test index ============================= #
dataset = LegacyTUDataset(args.dataset, raw_dir=args.dataset_path)
# add self loop. We add self loop for each graph here since the function "add_self_loop" does not
# support batch graph.
for i in range(len(dataset)):
dataset.graph_lists[i] = dgl.remove_self_loop(dataset.graph_lists[i])
dataset.graph_lists[i] = dgl.add_self_loop(dataset.graph_lists[i])
# preprocess: use node degree/label as node feature
if args.degree_as_feature:
dataset = degree_as_feature(dataset)
mode = "concat"
else:
mode = "replace"
dataset = node_label_as_feature(dataset, mode=mode)
num_training = int(len(dataset) * 0.9)
num_test = len(dataset) - num_training
train_set, test_set = random_split(dataset, [num_training, num_test])
train_loader = GraphDataLoader(train_set, batch_size=args.batch_size, shuffle=True, num_workers=1)
test_loader = GraphDataLoader(test_set, batch_size=args.batch_size, num_workers=1)
device = torch.device(args.device)
# Step 2: Create model =================================================================== #
num_feature, num_classes, _ = dataset.statistics()
args.in_dim = int(num_feature)
args.out_dim = int(num_classes)
args.edge_feat_dim = 0 # No edge feature in datasets that we use.
model = GraphClassifier(args).to(device)
# Step 3: Create training components ===================================================== #
optimizer = torch.optim.Adam(model.parameters(), lr=args.lr, amsgrad=True, weight_decay=args.weight_decay)
# Step 4: training epoches =============================================================== #
best_test_acc = 0.0
best_epoch = -1
train_times = []
for e in range(args.epochs):
s_time = time()
train_loss = train(model, optimizer, train_loader, device,
e, args.epochs)
train_times.append(time() - s_time)
test_acc = test(model, test_loader, device)
if test_acc > best_test_acc:
best_test_acc = test_acc
best_epoch = e + 1
if (e + 1) % args.print_every == 0:
log_format = "Epoch {}: loss={:.4f}, test_acc={:.4f}, best_test_acc={:.4f}"
print(log_format.format(e + 1, train_loss, test_acc, best_test_acc))
print("Best Epoch {}, final test acc {:.4f}".format(best_epoch, best_test_acc))
return best_test_acc, sum(train_times) / len(train_times)
def forward(self, graph, feat):
""""""
# add self_loop
graph = dgl.remove_self_loop(graph)
graph = dgl.add_self_loop(graph)
# prepare
x = feat
x = torch.mm(x, self.weight).view(-1, self.heads, self.out_channels)
out = self.propagate(graph, x)
return out
def forward(self, graph: dgl.DGLGraph, *__args, **__kwargs) -> _typing.Sequence[torch.Tensor]:
graph = dgl.remove_self_loop(graph)
graph = dgl.add_self_loop(graph)
h = graph.ndata['feat']
hidden_rep = [h]
for i in range(self.__num_layers):
h = self.__gcn_layers[i](graph, h)
h = self.__batch_normalizations[i](h)
h = torch.nn.functional.relu(h)
hidden_rep.append(h)
return hidden_rep
def state(self):
features = torch.stack([
x for i, x in enumerate(self.dataset.ndata['feat'])
if i in self.picked_nodes
])
features = np.array(features)
graph = dgl.add_self_loop(self.graph)
graph = dgl.to_networkx(graph).to_undirected()
self.embedding_model.fit(graph, features)
embedding = self.embedding_model.get_embedding()
embedding = np.sum(embedding, axis=0)
return embedding
def pass_data_iteratively(model, graphs, minibatch_size=64):
model.eval()
output = []
idx = np.arange(len(graphs))
for i in range(0, len(graphs), minibatch_size):
sampled_idx = idx[i:i + minibatch_size]
if len(sampled_idx) == 0:
continue
output.append(
model([dgl.add_self_loop(graphs[j][0])
for j in sampled_idx]).detach())
return torch.cat(output, 0)
def test_sg_conv(out_dim):
g = dgl.from_networkx(nx.erdos_renyi_graph(20, 0.3)).to(F.ctx())
g = dgl.add_self_loop(g)
ctx = F.ctx()
sgc = nn.SGConv(5, out_dim, 2)
sgc.initialize(ctx=ctx)
print(sgc)
# test #1: basic
h0 = F.randn((g.number_of_nodes(), 5))
h1 = sgc(g, h0)
assert h1.shape == (g.number_of_nodes(), out_dim)
