Graph neural networks (GNNs) combines superiority of both graph analytics and machine learning. GraphScope provides the capability to process learning tasks. In this tutorial, we demostrate how GraphScope trains a model with GCN.
The learning task is node classification on a citation network. In this task, the algorithm has to determine the label of the nodes in Cora dataset. The dataset consists of academic publications as the nodes and the citations between them as the links: if publication A cites publication B, then the graph has an edge from A to B. The nodes are classified into one of seven subjects, and our model will learn to predict this subject.
In this task, we use Graph Convolution Network (GCN) to train the model. The core of the GCN neural network model is a "graph convolution" layer. This layer is similar to a conventional dense layer, augmented by the graph adjacency matrix to use information about a node's connections.
This tutorial has the following steps:
# Install graphscope package if you are NOT in the Playground
!pip3 install graphscope
!pip3 uninstall -y importlib_metadata # Address an module conflict issue on colab.google. Remove this line if you are not on colab.
# Import the graphscope module.
import graphscope
graphscope.set_option(show_log=False) # enable logging
# Load cora dataset
from graphscope.dataset import load_cora
graph = load_cora()
Then, we need to define a feature list for training. The training feature list should be seleted from the vertex properties. In this case, we choose all the properties prefix with "feat_" as the training features.
With the featrue list, next we launch a learning engine with the graphlearn method of graphscope.
In this case, we specify the GCN training over "paper" nodes and "cites" edges.
With "gen_labels", we split the "paper" nodes into three parts, 75% are used as training set, 10% are used for validation and 15% used for testing.
# define the features for learning
paper_features = []
for i in range(1433):
paper_features.append("feat_" + str(i))
# launch a learning engine.
lg = graphscope.graphlearn(
graph,
nodes=[("paper", paper_features)],
edges=[("paper", "cites", "paper")],
gen_labels=[
("train", "paper", 100, (0, 75)),
("val", "paper", 100, (75, 85)),
("test", "paper", 100, (85, 100)),
],
)
We use the builtin GCN model to define the training process. You can find more detail about all the builtin learning models on Graph Learning Model
In the example, we use tensorflow as "NN" backend trainer.
from graphscope.learning.examples import GCN
from graphscope.learning.graphlearn.python.model.tf.optimizer import get_tf_optimizer
from graphscope.learning.graphlearn.python.model.tf.trainer import LocalTFTrainer
# supervised GCN.
def train(config, graph):
def model_fn():
return GCN(
graph,
config["class_num"],
config["features_num"],
config["batch_size"],
val_batch_size=config["val_batch_size"],
test_batch_size=config["test_batch_size"],
categorical_attrs_desc=config["categorical_attrs_desc"],
hidden_dim=config["hidden_dim"],
in_drop_rate=config["in_drop_rate"],
neighs_num=config["neighs_num"],
hops_num=config["hops_num"],
node_type=config["node_type"],
edge_type=config["edge_type"],
full_graph_mode=config["full_graph_mode"],
)
trainer = LocalTFTrainer(
model_fn,
epoch=config["epoch"],
optimizer=get_tf_optimizer(
config["learning_algo"], config["learning_rate"], config["weight_decay"]
),
)
trainer.train_and_evaluate()
# define hyperparameters
config = {
"class_num": 7, # output dimension
"features_num": 1433,
"batch_size": 140,
"val_batch_size": 300,
"test_batch_size": 1000,
"categorical_attrs_desc": "",
"hidden_dim": 128,
"in_drop_rate": 0.5,
"hops_num": 2,
"neighs_num": [5, 5],
"full_graph_mode": False,
"agg_type": "gcn", # mean, sum
"learning_algo": "adam",
"learning_rate": 0.01,
"weight_decay": 0.0005,
"epoch": 5,
"node_type": "paper",
"edge_type": "cites",
}
After define training process and hyperparameters,
Now we can start the traning process with learning engine "" and the hyperparameters configurations.
train(config, lg)