大规模图数据思路代码_neighborsampler+gcn-CSDN博客

本文链接：https://blog.csdn.net/L_goodboy/article/details/140738160

GCN+邻居采样器

import torch
import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
from torch_geometric.nn import GCNConv
from torch_geometric.loader import NeighborSampler

# 定义图神经网络模型
class GCN(torch.nn.Module):
    def __init__(self, num_features, num_classes):
        super(GCN, self).__init__()
        self.conv1 = GCNConv(num_features, 16)
        self.conv2 = GCNConv(16, num_classes)

    def forward(self, x, edge_index, size):
        # 使用邻居信息进行采样
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index, size)
        return F.log_softmax(x, dim=1)


# 加载数据集
dataset = Planetoid(root='/tmp/Cora', name='Cora')
data = dataset[0]

# 创建邻居采样迭代器
train_loader = NeighborSampler(data.edge_index, 
                                node_idx=data.train_mask, 
                                sizes=[10, 10],  # 每层采样的邻居数
                                batch_size=64,   # 每次批次的节点数
                                num_nodes=data.num_nodes)

# 初始化模型和优化器
model = GCN(num_features=dataset.num_node_features, num_classes=dataset.num_classes)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model.to(device)

# 训练模型
def train(loader, model, optimizer):
    model.train()
    total_loss = 0
    for batch_size, n_id, adj in loader:
        optimizer.zero_grad()
        # 通过采样得到节点特征和边列表
        x = data.x[n_id].to(device)
        out = model(x, adj.edge_index, adj.size)
        loss = F.nll_loss(out[adj.node_idx], data.y[adj.node_idx].to(device))
        loss.backward()
        optimizer.step()
        total_loss += loss.item()
    return total_loss

# 评估模型
def test(model):
    model.eval()
    with torch.no_grad():
        out = model(data.x.to(device), data.edge_index.to(device), data.num_nodes)
    pred = out.argmax(dim=1)
    correct = pred[data.test_mask.to(device)] == data.y[data.test_mask.to(device)]
    acc = int(correct.sum()) / int(data.test_mask.sum())
    return acc

# 训练和验证
for epoch in range(200):
    loss = train(train_loader, model, optimizer)
    if epoch % 20 == 0:
        train_acc = test(model)
        print(f'Epoch {epoch}, Loss: {loss:.4f}, Train Accuracy: {train_acc:.4f}')

# 测试模型
test_acc = test(model)
print(f'Test Accuracy: {test_acc:.4f}')

GCN+邻居采样器

import torch
import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
from torch_geometric.nn import GCNConv
from torch_geometric.data import DataLoader
from torch_geometric.loader import NeighborSampler

class GCN(torch.nn.Module):
    def __init__(self, num_features, num_classes):
        super(GCN, self).__init__()
        self.conv1 = GCNConv(num_features, 16)
        self.conv2 = GCNConv(16, num_classes)

    def forward(self, x, edge_index):
        # 第一层图卷积
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        # 第二层图卷积
        x = self.conv2(x, edge_index)
        return F.log_softmax(x, dim=1)

# 载入数据集（Cora）
dataset = Planetoid(root='/tmp/Cora', name='Cora')
data = dataset[0]

# 定义邻居采样器
def create_neighbor_sampler(data, num_neighbors):
    return NeighborSampler(
        data.edge_index, 
        sizes=num_neighbors,  # 每层采样的邻居数目
        batch_size=None,      # 批量大小
        shuffle=True,         # 是否打乱数据
        num_workers=0         # 工作线程数量
    )

num_neighbors = [10, 5]  # 第一层采样10个邻居，第二层采样5个邻居
neighbor_sampler = create_neighbor_sampler(data, num_neighbors)

# 模型和优化器初始化
model = GCN(num_features=dataset.num_node_features, num_classes=dataset.num_classes)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = model.to(device)
data = data.to(device)

# 训练函数
def train():
    model.train()
    total_loss = 0
    for batch_size, n_id, adjs in neighbor_sampler:
        adjs = [adj.to(device) for adj in adjs]
        out = model(data.x[n_id], adjs[0].edge_index)
        loss = F.nll_loss(out[data.train_mask[n_id]], data.y[data.train_mask[n_id]])
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        total_loss += loss.item()
    return total_loss

# 测试函数
def test():
    model.eval()
    with torch.no_grad():
        out = model(data.x, data.edge_index)
    pred = out.argmax(dim=1)
    correct = pred[data.test_mask] == data.y[data.test_mask]
    acc = int(correct.sum()) / int(data.test_mask.sum())
    return acc

# 训练和验证过程
for epoch in range(200):
    loss = train()
    if epoch % 20 == 0:
        train_acc = test()
        print(f'Epoch {epoch}, Loss: {loss:.4f}, Train Accuracy: {train_acc:.4f}')

# 测试模型
test_acc = test()
print(f'Test Accuracy: {test_acc:.4f}')

GraphSAGE+邻居采样器

import torch
import torch.nn.functional as F
from torch_geometric.data import Planetoid
from torch_geometric.nn import SAGEConv
from torch_geometric.loader import NeighborSampler
from torch_geometric.utils import to_undirected
from torch_geometric.transforms import ToUndirected

# 1. 数据加载
dataset = Planetoid(root='/tmp/Cora', name='Cora', transform=ToUndirected())
data = dataset[0]

# 2. GraphSAGE 模型定义
class GraphSAGE(torch.nn.Module):
    def __init__(self, in_channels, hidden_channels, out_channels):
        super(GraphSAGE, self).__init__()
        self.conv1 = SAGEConv(in_channels, hidden_channels)
        self.conv2 = SAGEConv(hidden_channels, out_channels)
    
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = self.conv2(x, edge_index)
        return x

# 3. 采样器定义
def get_loader(data, num_neighbors):
    edge_index = data.edge_index
    sampler = NeighborSampler(edge_index, sizes=num_neighbors, batch_size=64, shuffle=True)
    return sampler

# 4. 训练和评估
def train(model, loader, optimizer, data):
    model.train()
    total_loss = 0
    for batch_size, n_id, adj in loader:
        adj = adj.to(data.edge_index.device)
        n_id = n_id.to(data.x.device)
        out = model(data.x[n_id], adj.edge_index)
        loss = F.cross_entropy(out[data.train_mask], data.y[data.train_mask])
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        total_loss += loss.item()
    return total_loss / len(loader)

def test(model, data):
    model.eval()
    with torch.no_grad():
        out = model(data.x, data.edge_index)
        pred = out.argmax(dim=1)
        test_acc = (pred[data.test_mask] == data.y[data.test_mask]).sum().item() / data.test_mask.sum().item()
        return test_acc

# 实例化模型
model = GraphSAGE(dataset.num_features, 64, dataset.num_classes)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

# 采样器
loader = get_loader(data, num_neighbors=[10, 10])

# 训练
for epoch in range(1, 101):
    loss = train(model, loader, optimizer, data)
    acc = test(model, data)
    print(f'Epoch: {epoch}, Loss: {loss:.4f}, Test Accuracy: {acc:.4f}')

GAT+邻居采样器

import torch
import torch.nn.functional as F
from torch_geometric.data import Planetoid
from torch_geometric.nn import GATConv
from torch_geometric.loader import NeighborSampler
from torch_geometric.transforms import ToUndirected

# 1. 数据加载
dataset = Planetoid(root='/tmp/Cora', name='Cora', transform=ToUndirected())
data = dataset[0]

# 2. GAT 模型定义
class GAT(torch.nn.Module):
    def __init__(self, in_channels, hidden_channels, out_channels, num_heads):
        super(GAT, self).__init__()
        self.conv1 = GATConv(in_channels, hidden_channels, heads=num_heads, concat=True)
        self.conv2 = GATConv(hidden_channels * num_heads, out_channels, heads=1, concat=False)
    
    def forward(self, x, edge_index):
        x = self.conv1(x, edge_index)
        x = F.elu(x)
        x = self.conv2(x, edge_index)
        return x

# 3. 采样器定义
def get_loader(data, num_neighbors):
    edge_index = data.edge_index
    sampler = NeighborSampler(edge_index, sizes=num_neighbors, batch_size=64, shuffle=True)
    return sampler

# 4. 训练和评估
def train(model, loader, optimizer, data):
    model.train()
    total_loss = 0
    for batch_size, n_id, adj in loader:
        adj = adj.to(data.edge_index.device)
        n_id = n_id.to(data.x.device)
        out = model(data.x[n_id], adj.edge_index)
        loss = F.cross_entropy(out[data.train_mask], data.y[data.train_mask])
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        total_loss += loss.item()
    return total_loss / len(loader)

def test(model, data):
    model.eval()
    with torch.no_grad():
        out = model(data.x, data.edge_index)
        pred = out.argmax(dim=1)
        test_acc = (pred[data.test_mask] == data.y[data.test_mask]).sum().item() / data.test_mask.sum().item()
        return test_acc

# 实例化模型
model = GAT(dataset.num_features, 8, dataset.num_classes, num_heads=8)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01)

# 采样器
loader = get_loader(data, num_neighbors=[15, 10, 5])

# 训练
for epoch in range(1, 101):
    loss = train(model, loader, optimizer, data)
    acc = test(model, data)
    print(f'Epoch: {epoch}, Loss: {loss:.4f}, Test Accuracy: {acc:.4f}')