# 導(dǎo)入模型數(shù)據(jù)集
from torch_geometric.datasets import ModelNet
# 加載ModelNet數(shù)據(jù)集
dataset = ModelNet(root='./modelnet', name='10')
data = dataset[0]
# 定義超級(jí)參數(shù)
num_points = 1024
batch_size = 32
train_dataset_size = 8000
# 將數(shù)據(jù)集分割成訓(xùn)練、驗(yàn)證及測(cè)試三個(gè)數(shù)據(jù)集
train_dataset = data[0:train_dataset_size]
val_dataset = data[train_dataset_size: 9000]
test_dataset = data[9000:]
# 定義數(shù)據(jù)加載批處理器
train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True)
val_loader = DataLoader(val_dataset, batch_size=batch_size, shuffle=False)
test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False)

from torch.nn import Sequential as Seq, Linear as Lin, ReLU
from torch_geometric.nn import EdgeConv, global_max_pool
class EdgeCNN(torch.nn.Module):
    def __init__(self, dataset):
        super(EdgeCNN, self).__init__()
        # 定義基礎(chǔ)參數(shù)
        self.input_dim = dataset.num_features
        self.output_dim = dataset.num_classes
        self.num_points = num_points
        # 定義模型結(jié)構(gòu)
        self.conv1 = EdgeConv(Seq(Lin(self.input_dim, 32), ReLU()))
        self.conv2 = EdgeConv(Seq(Lin(32, 64), ReLU()))
        self.conv3 = EdgeConv(Seq(Lin(64, 128), ReLU()))
        self.conv4 = EdgeConv(Seq(Lin(128, 256), ReLU()))
        self.fc1 = torch.nn.Linear(256, 1024)
        self.fc2 = torch.nn.Linear(1024, self.output_dim)
    def forward(self, pos, batch):
        # 構(gòu)造圖
        edge_index = radius_graph(pos, r=0.6, batch=batch, loop=False)
        # 第一層CNN模型的卷積 + 池化處理
        x = F.relu(self.conv1(x=pos, edge_index=edge_index))
        x = global_max_pool(x, batch)
        # 第二層CNN模型的卷積 + 池化處理
        edge_index = radius_graph(x, r=0.9, batch=batch, loop=False)
        x = F.relu(self.conv2(x=x, edge_index=edge_index))
        x = global_max_pool(x, batch)
        # 第三層CNN模型的卷積 + 池化處理
        edge_index = radius_graph(x, r=1.2, batch=batch, loop=False)
        x = F.relu(self.conv3(x=x, edge_index=edge_index))
        x = global_max_pool(x, batch)
        # 第四層CNN模型的卷積 + 池化處理
        edge_index = radius_graph(x, r=1.5, batch=batch, loop=False)
        x = F.relu(self.conv4(x=x, edge_index=edge_index))
        # 定義全連接網(wǎng)絡(luò)
        x = global_max_pool(x, batch)
        x = F.relu(self.fc1(x))
        x = self.fc2(x)
        return F.log_softmax(x, dim=-1)

# 定義訓(xùn)練計(jì)劃，包括損失函數(shù)、優(yōu)化器及迭代次數(shù)等
train_epochs = 50
learning_rate = 0.01
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(edge_cnn.parameters(), lr=learning_rate)
losses_per_epoch = []
accuracies_per_epoch = []
for epoch in range(train_epochs):
    running_loss = 0.0
    running_corrects = 0.0
    count = 0.0
    for samples in train_loader:
        optimizer.zero_grad()
        pos, batch, label = samples.pos, samples.batch, samples.y.to(torch.long)
        out = edge_cnn(pos, batch)
        loss = criterion(out, label)
        loss.backward()
        optimizer.step()
        running_loss += loss.item() / len(train_dataset)
        running_corrects += torch.sum(torch.argmax(out, dim=1) == label).item() / len(train_dataset)
        count += 1
    losses_per_epoch.append(running_loss)
    accuracies_per_epoch.append(running_corrects)
    if (epoch + 1) % 5 == 0:
        print("Train Epoch {}/{} Loss {:.4f} Accuracy {:.4f}".format(
            epoch + 1, train_epochs, running_loss, running_corrects))