def __init__(self, app: str, model: str, **kwargs):
super(RecommendationPipepline, self).__init__(app, model=model, **kwargs)
if "data" in kwargs:
data = kwargs["data"]
val_data = test_data = data[-100:, :]
data = build_recommendation_data("custom", data, val_data, test_data)
self.data_path = kwargs.get("data_path", "tmp_data.pt")
self.batch_size = kwargs.get("batch_size", 128)
torch.save(data, self.data_path)
self.dataset = NodeDataset(path=self.data_path, scale_feat=False)
elif "dataset" in kwargs:
dataset = kwargs.pop("dataset")
self.dataset = build_dataset_from_name(dataset)
else:
print("Please provide recommendation data!")
exit(0)
self.batch_size = kwargs.get("batch_size", 2048)
self.n_items = self.dataset[0].n_params["n_items"]
args = get_default_args(task="recommendation", dataset="ali", model=model, **kwargs)
args.model = args.model[0]
# task = build_task(args, dataset=self.dataset)
# task.train()
# self.model = task.model
self.model = build_model(args)
self.model.eval()
self.user_emb, self.item_emb = self.model.generate()
def get_default_args_kg(dataset,
model,
dw="gnn_kg_link_prediction_dw",
mw="gnn_kg_link_prediction_mw"):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict_kg.items():
args.__setattr__(key, value)
return args
def get_default_args_graph_clf(dataset,
model,
dw="graph_classification_dw",
mw="graph_classification_mw"):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict.items():
args.__setattr__(key, value)
return args
def __init__(self, app: str, model: str, **kwargs):
super(GenerateEmbeddingPipeline, self).__init__(app,
model=model,
**kwargs)
match_path = os.path.join(os.path.dirname(os.path.realpath(__file__)),
"match.yml")
with open(match_path, "r", encoding="utf8") as f:
match = yaml.load(f, Loader=yaml.FullLoader)
objective = match.get("unsupervised_node_classification", None)
for pair_dict in objective:
if "blogcatalog" in pair_dict["dataset"]:
emb_models = pair_dict["model"]
elif "cora" in pair_dict["dataset"]:
gnn_models = pair_dict["model"]
if model in emb_models:
self.method_type = "emb"
args = get_default_args(task="unsupervised_node_classification",
dataset="blogcatalog",
model=model,
**kwargs)
elif model in gnn_models:
self.method_type = "gnn"
args = get_default_args(task="unsupervised_node_classification",
dataset="cora",
model=model,
**kwargs)
else:
print("Please choose a model from ", emb_models, "or", gnn_models)
exit(0)
self.data_path = kwargs.get("data_path", "tmp_data.pt")
self.num_features = kwargs.get("num_features", None)
if self.num_features is not None:
args.num_features = self.num_features
elif self.method_type == "gnn":
print("Please provide num_features for gnn model!")
exit(0)
args.model = args.model[0]
self.model = build_model(args)
self.trainer = self.model.get_trainer(self.model, args)
if self.trainer is not None:
self.trainer = self.trainer(args)
def get_default_args_ne(dataset,
model,
dw="network_embedding_dw",
mw="network_embedding_mw"):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict.items():
args.__setattr__(key, value)
return args
def get_default_args_for_unsup_nn(dataset,
model,
dw="node_classification_dw",
mw="self_auxiliary_mw"):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict.items():
args.__setattr__(key, value)
return args
def get_default_args_emb(dataset,
model,
dw="heterogeneous_embedding_dw",
mw="heterogeneous_embedding_mw"):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict_emb.items():
args.__setattr__(key, value)
return args
def test_train():
args = get_default_args(dataset="cora", model="gcn", epochs=10, cpu=True)
args.dataset = args.dataset[0]
args.model = args.model[0]
args.seed = args.seed[0]
result = train(args)
assert "test_acc" in result
assert result["test_acc"] > 0
def test_get_default_args():
args = options.get_default_args(task="node_classification",
dataset="cora",
model="gcn")
assert args.task == "node_classification"
assert args.model == "gcn"
assert args.dataset == "cora"
assert args.hidden_size > 0
assert args.lr > 0
def test_get_default_args():
args = options.get_default_args(dataset=["cora", "citeseer"],
model=["gcn", "gat"],
hidden_size=128)
assert args.model[0] == "gcn"
assert args.model[1] == "gat"
assert args.dataset[0] == "cora"
assert args.dataset[1] == "citeseer"
assert args.hidden_size == 128
def get_default_args_generative(dataset,
model,
dw="node_classification_dw",
mw="self_auxiliary_mw",
**kwargs):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict.items():
args.__setattr__(key, value)
for key, value in kwargs.items():
args.__setattr__(key, value)
return args
def test_set_best_config():
args = get_default_args(task="node_classification",
dataset="citeseer",
model="gat")
args.model = args.model[0]
args.dataset = args.dataset[0]
args = set_best_config(args)
assert args.lr == 0.005
assert args.max_epoch == 1000
assert args.weight_decay == 0.001
def __init__(self, app: str, model: str, **kwargs):
super(GenerateEmbeddingPipeline, self).__init__(app, model=model, **kwargs)
self.kwargs = kwargs
emb_models = [
"prone",
"netmf",
"netsmf",
"deepwalk",
"line",
"node2vec",
"hope",
"sdne",
"grarep",
"dngr",
"spectral",
]
gnn_models = ["dgi", "mvgrl", "grace", "unsup_graphsage"]
if model in emb_models:
self.method_type = "emb"
args = get_default_args(dataset="blogcatalog", model=model, **kwargs)
elif model in gnn_models:
self.method_type = "gnn"
args = get_default_args(dataset="cora", model=model, **kwargs)
else:
print("Please choose a model from ", emb_models, "or", gnn_models)
exit(0)
self.data_path = kwargs.get("data_path", "tmp_data.pt")
self.num_features = kwargs.get("num_features", None)
if self.num_features is not None:
args.num_features = self.num_features
elif self.method_type == "gnn":
print("Please provide num_features for gnn model!")
exit(0)
args.model = args.model[0]
self.args = args
def test_train():
args = get_default_args(task="node_classification",
dataset="cora",
model="gcn",
max_epoch=10,
cpu=True)
args.dataset = args.dataset[0]
args.model = args.model[0]
args.seed = args.seed[0]
result = train(args)
assert "Acc" in result
assert result["Acc"] > 0
def raw_experiment(task: str, dataset, model, **kwargs):
if "args" not in kwargs:
args = get_default_args(task=task, dataset=dataset, model=model, **kwargs)
else:
args = kwargs["args"]
variants = list(gen_variants(dataset=args.dataset, model=args.model, seed=args.seed))
variants = check_task_dataset_model_match(task, variants)
results_dict = defaultdict(list)
results = [train(args) for args in variant_args_generator(args, variants)]
for variant, result in zip(variants, results):
results_dict[variant[:-1]].append(result)
tablefmt = kwargs["tablefmt"] if "tablefmt" in kwargs else "github"
output_results(results_dict, tablefmt)
return results_dict
def experiment(dataset, model=None, **kwargs):
if model is None:
model = "autognn"
if isinstance(dataset, str) or isinstance(dataset, Dataset):
dataset = [dataset]
if isinstance(model, str) or isinstance(model, nn.Module):
model = [model]
if "args" not in kwargs:
args = get_default_args(dataset=[str(x) for x in dataset],
model=[str(x) for x in model],
**kwargs)
else:
args = kwargs["args"]
for key, value in kwargs.items():
if key != "args":
args.__setattr__(key, value)
if isinstance(model[0], nn.Module):
args.model = [x.model_name for x in model]
print(args)
args.dataset = dataset
args.model = model
if args.max_epoch is not None:
warnings.warn(
"The max_epoch is deprecated and will be removed in the future, please use epochs instead!"
)
args.epochs = args.max_epoch
if len(model) == 1 and isinstance(model[0], str) and model[0] == "autognn":
if not hasattr(args, "search_space"):
args.search_space = default_search_space
if not hasattr(args, "seed"):
args.seed = [1, 2]
if not hasattr(args, "n_trials"):
args.n_trials = 20
if hasattr(args, "search_space"):
return auto_experiment(args)
return raw_experiment(args)
.format(acc))
# %%
# Experiment API
# --------------
# CogDL在训练上提供了更易于使用的 API ，即Experiment
from cogdl import experiment
experiment(model="gcn", dataset="cora")
#或者，您可以单独创建每个组件并使用CogDL 中的 build_dataset , build_model 来手动运行该过程。
from cogdl import experiment
from cogdl.datasets import build_dataset
from cogdl.models import build_model
from cogdl.options import get_default_args
args = get_default_args(model="gcn", dataset="cora")
dataset = build_dataset(args)
model = build_model(args)
experiment(model=model, dataset=dataset)
# %%
# 如何保存训练好的模型？
# --------------------------
experiment(model="gcn", dataset="cora", checkpoint_path="gcn_cora.pt")
# 当训练停止时，模型将保存在 gcn_cora.pt 中。如果你想从之前的checkpoint继续训练，使用不同的参数（如学习率、权重衰减等），保持相同的模型参数（如hidden size、模型层数），可以像下面这样做：
experiment(model="gcn",
dataset="cora",
checkpoint_path="gcn_cora.pt",
resume_training=True)
# %%
def get_default_args_agc(dataset, model, dw=None, mw=None):
args = get_default_args(dataset=dataset, model=model, dw=dw, mw=mw)
for key, value in default_dict.items():
args.__setattr__(key, value)
return args
from cogdl.tasks import build_task
from cogdl.options import get_default_args
args = get_default_args(task="node_classification",
dataset="cora",
model="gcn")
task = build_task(args)
ret = task.train()
