# 通过调用paddle.io.DataLoader来构造reader，这里需要使用DistributedBatchSampler为多张卡拆分数据
train_sampler = paddle.io.DistributedBatchSampler(MnistDataset(mode='train'),
batch_size=BATCH_SIZE,
drop_last=True)
train_reader = paddle.io.DataLoader(MnistDataset(mode='train'),
batch_sampler=train_sampler)
for epoch in range(epoch_num):
for batch_id, data in enumerate(train_reader()):
img = data[0]
label = data[1]
label.stop_gradient = True
# 网络正向执行
pred, acc = mnist(img, label)
# 计算损失值
loss = paddle.nn.functional.cross_entropy(pred, label)
avg_loss = paddle.mean(loss)
avg_loss.backward()
# 参数更新
adam.step()
# 将本次计算的梯度值清零，以便进行下一次迭代和梯度更新
adam.clear_grad()
# 输出对应epoch、batch_id下的损失值
if batch_id % 100 == 0 and batch_id is not 0:
print("Epoch {} step {}, Loss = {:}, Accuracy = {:}".format(
epoch, batch_id, avg_loss.numpy(), acc))
def main(args):
"""
Model training for one epoch and return the average loss and model evaluating to monitor pcc.
"""
paddle.set_device('gpu:{}'.format(args.device) if args.use_cuda else 'cpu')
logging.info('Load data ...')
dataset = InMemoryDataset(npz_data_path=args.data_path)
train_ds = Dataset(dataset[1])
test_ds = Dataset(dataset[0])
train_loader = train_ds.get_data_loader(batch_size=args.batch_size,
collate_fn=collate_fn)
test_loader = test_ds.get_data_loader(batch_size=args.batch_size,
collate_fn=collate_fn)
logging.info("Data loaded.")
model = CDRModel(args)
optim = Adam(learning_rate=args.lr, parameters=model.parameters())
criterion = nn.MSELoss()
global_step = 0
best_pcc = 0.0
os.makedirs(args.output_path, exist_ok=True)
best_model = os.path.join(args.output_path, 'best_model.pdparams')
for epoch in range(1, args.epoch_num + 1):
model.train()
for idx, batch_data in enumerate(train_loader):
graphs, mut, gexpr, met, label = batch_data
g = pgl.Graph.batch(graphs).tensor()
mut = paddle.to_tensor(mut)
gexpr = paddle.to_tensor(gexpr)
met = paddle.to_tensor(met)
label = paddle.to_tensor(label)
pred = model([g, mut, gexpr, met])
train_loss = paddle.pow(criterion(pred[:, 0], label)[0], 0.5)
train_loss.backward()
train_pcc = pearsonr(pred[:, 0].numpy(), label.numpy())[0]
optim.step()
optim.clear_grad()
global_step += 1
if global_step % 500 == 0:
message = "train: epoch %d | step %d | " % (epoch, global_step)
message += "loss %.6f ｜ pcc %.4f" % (train_loss, train_pcc)
log.info(message)
result = evaluate(model, test_loader, criterion)
message = "eval: epoch %d | step %d " % (epoch, global_step)
for key, value in result.items():
message += "| %s %.6f" % (key, value)
log.info(message)
if best_pcc < result['pcc']:
best_pcc = result['pcc']
paddle.save(model.state_dict(), best_model)
log.info("best evaluating accuracy: %.6f" % best_pcc)
def main(args):
ds = GINDataset(args.data_path,
args.dataset_name,
self_loop=not args.train_eps,
degree_as_nlabel=True)
args.feat_size = ds.dim_nfeats
train_ds, test_ds = fold10_split(ds,
fold_idx=args.fold_idx,
seed=args.seed)
train_loader = Dataloader(train_ds,
batch_size=args.batch_size,
shuffle=True,
num_workers=1,
collate_fn=collate_fn)
test_loader = Dataloader(test_ds,
batch_size=args.batch_size,
shuffle=False,
num_workers=1,
collate_fn=collate_fn)
model = GINModel(args, ds.gclasses)
epoch_step = len(train_loader)
boundaries = [
i for i in range(50 * epoch_step, args.epochs *
epoch_step, epoch_step * 50)
]
values = [args.lr * 0.5**i for i in range(0, len(boundaries) + 1)]
scheduler = paddle.optimizer.lr.PiecewiseDecay(boundaries=boundaries,
values=values,
verbose=False)
optim = Adam(learning_rate=scheduler, parameters=model.parameters())
criterion = nn.loss.CrossEntropyLoss()
global_step = 0
best_acc = 0.0
for epoch in range(1, args.epochs + 1):
model.train()
for idx, batch_data in enumerate(train_loader):
graphs, labels = batch_data
g = pgl.Graph.batch(graphs).tensor()
labels = paddle.to_tensor(labels)
pred = model(g)
train_loss = criterion(pred, labels)
train_loss.backward()
train_acc = paddle.metric.accuracy(input=pred, label=labels, k=1)
optim.step()
optim.clear_grad()
scheduler.step()
global_step += 1
if global_step % 10 == 0:
message = "train: epoch %d | step %d | " % (epoch, global_step)
message += "loss %.6f | acc %.4f" % (train_loss, train_acc)
log.info(message)
result = evaluate(model, test_loader, criterion)
message = "eval: epoch %d | step %d | " % (epoch, global_step)
for key, value in result.items():
message += " | %s %.6f" % (key, value)
log.info(message)
if best_acc < result['acc']:
best_acc = result['acc']
log.info("best evaluating accuracy: %.6f" % best_acc)
def main(config):
if dist.get_world_size() > 1:
dist.init_parallel_env()
if dist.get_rank() == 0:
timestamp = datetime.now().strftime("%Hh%Mm%Ss")
log_path = os.path.join(config.log_dir,
"tensorboard_log_%s" % timestamp)
writer = SummaryWriter(log_path)
log.info("loading data")
raw_dataset = GraphPropPredDataset(name=config.dataset_name)
config.num_class = raw_dataset.num_tasks
config.eval_metric = raw_dataset.eval_metric
config.task_type = raw_dataset.task_type
mol_dataset = MolDataset(config,
raw_dataset,
transform=make_multihop_edges)
splitted_index = raw_dataset.get_idx_split()
train_ds = Subset(mol_dataset, splitted_index['train'], mode='train')
valid_ds = Subset(mol_dataset, splitted_index['valid'], mode="valid")
test_ds = Subset(mol_dataset, splitted_index['test'], mode="test")
log.info("Train Examples: %s" % len(train_ds))
log.info("Val Examples: %s" % len(valid_ds))
log.info("Test Examples: %s" % len(test_ds))
fn = CollateFn(config)
train_loader = Dataloader(train_ds,
batch_size=config.batch_size,
shuffle=True,
num_workers=config.num_workers,
collate_fn=fn)
valid_loader = Dataloader(valid_ds,
batch_size=config.batch_size,
num_workers=config.num_workers,
collate_fn=fn)
test_loader = Dataloader(test_ds,
batch_size=config.batch_size,
num_workers=config.num_workers,
collate_fn=fn)
model = ClassifierNetwork(config.hidden_size, config.out_dim,
config.num_layers, config.dropout_prob,
config.virt_node, config.K, config.conv_type,
config.appnp_hop, config.alpha)
model = paddle.DataParallel(model)
optim = Adam(learning_rate=config.lr, parameters=model.parameters())
criterion = nn.loss.BCEWithLogitsLoss()
evaluator = Evaluator(config.dataset_name)
best_valid = 0
global_step = 0
for epoch in range(1, config.epochs + 1):
model.train()
for idx, batch_data in enumerate(train_loader):
g, mh_graphs, labels, unmask = batch_data
g = g.tensor()
multihop_graphs = []
for item in mh_graphs:
multihop_graphs.append(item.tensor())
g.multi_hop_graphs = multihop_graphs
labels = paddle.to_tensor(labels)
unmask = paddle.to_tensor(unmask)
pred = model(g)
pred = paddle.masked_select(pred, unmask)
labels = paddle.masked_select(labels, unmask)
train_loss = criterion(pred, labels)
train_loss.backward()
optim.step()
optim.clear_grad()
if global_step % 80 == 0:
message = "train: epoch %d | step %d | " % (epoch, global_step)
message += "loss %.6f" % (train_loss.numpy())
log.info(message)
if dist.get_rank() == 0:
writer.add_scalar("loss", train_loss.numpy(), global_step)
global_step += 1
valid_result = evaluate(model, valid_loader, criterion, evaluator)
message = "valid: epoch %d | step %d | " % (epoch, global_step)
for key, value in valid_result.items():
message += " | %s %.6f" % (key, value)
if dist.get_rank() == 0:
writer.add_scalar("valid_%s" % key, value, global_step)
log.info(message)
test_result = evaluate(model, test_loader, criterion, evaluator)
message = "test: epoch %d | step %d | " % (epoch, global_step)
for key, value in test_result.items():
message += " | %s %.6f" % (key, value)
if dist.get_rank() == 0:
writer.add_scalar("test_%s" % key, value, global_step)
log.info(message)
if best_valid < valid_result[config.metrics]:
best_valid = valid_result[config.metrics]
best_valid_result = valid_result
best_test_result = test_result
message = "best result: epoch %d | " % (epoch)
message += "valid %s: %.6f | " % (config.metrics,
best_valid_result[config.metrics])
message += "test %s: %.6f | " % (config.metrics,
best_test_result[config.metrics])
log.info(message)
message = "final eval best result:%.6f" % best_valid_result[config.metrics]
log.info(message)
message = "final test best result:%.6f" % best_test_result[config.metrics]
log.info(message)