PyTorch怎么安装和使用

$ python -c "import torch; print(torch.__version__)">>> 1.2.0

$ pip install --verbose --no-cache-dir torch-scatter$ pip install --verbose --no-cache-dir torch-sparse$ pip install --verbose --no-cache-dir torch-cluster$ pip install --verbose --no-cache-dir torch-spline-conv (optional)$ pip install torch-geometric

def spawn(self, cmd):    spawn(cmd, dry_run=self.dry_run)

def spawn(self, cmd):    spawn(cmd, dry_run=self.dry_run)

import torchfrom torch_geometric.data import Data#边，shape = [2,num_edge]edge_index = torch.tensor([[0, 1, 1, 2],                           [1, 0, 2, 1]], dtype=torch.long)#点，shape = [num_nodes, num_node_features]x = torch.tensor([[-1], [0], [1]], dtype=torch.float)data = Data(x=x, edge_index=edge_index)>>> Data(edge_index=[2, 4], x=[3, 1])

from torch_geometric.datasets import TUDatasetdataset = TUDataset(root='/tmp/ENZYMES', name='ENZYMES')>>> ENZYMES(600)len(dataset)>>> 600dataset.num_classes>>> 6dataset.num_node_features>>> 3

data = dataset[0]>>> Data(edge_index=[2, 168], x=[37, 3], y=[1])#可以看出这个图包含边168/2=84条，节点37个，每个节点包含三个特征data.is_undirected()>>> True

from torch_geometric.datasets import Planetoiddataset = Planetoid(root='/tmp/Cora', name='Cora')>>> Cora()#可以看到这个数据集只有一个图len(dataset)>>> 1dataset.num_classes>>> 7dataset.num_node_features>>> 1433#train_maskdata = dataset[0]>>> Data(edge_index=[2, 10556], test_mask=[2708],         train_mask=[2708], val_mask=[2708], x=[2708, 1433], y=[2708])#用来训练的数据量data.train_mask.sum().item()>>> 140#用来验证的数据量data.val_mask.sum().item()>>> 500#用来测试的数据量data.test_mask.sum().item()>>> 1000

import torchimport torchimport torch.nn.functional as Ffrom torch_geometric.nn import GCNConv#数据集加载from torch_geometric.datasets import Planetoiddataset = Planetoid(root='/tmp/Cora', name='Cora')#网络定义class Net(torch.nn.Module):    def __init__(self):        super(Net, self).__init__()        self.conv1 = GCNConv(dataset.num_node_features, 16)        self.conv2 = GCNConv(16, dataset.num_classes)    def forward(self, data):        x, edge_index = data.x, data.edge_index        x = self.conv1(x, edge_index)        x = F.relu(x)        x = F.dropout(x, training=self.training)        x = self.conv2(x, edge_index)        return F.log_softmax(x, dim=1)device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')model = Net().to(device)data = dataset[0].to(device)optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)#网络训练model.train()for epoch in range(200):    optimizer.zero_grad()    out = model(data)    loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])    loss.backward()    optimizer.step()#测试model.eval()_, pred = model(data).max(dim=1)correct = float(pred[data.test_mask].eq(data.y[data.test_mask]).sum().item())acc = correct / data.test_mask.sum().item()print('Accuracy: {:.4f}'.format(acc))

import torchfrom torch_geometric.nn import MessagePassingfrom torch_geometric.utils import add_self_loops, degreeclass GCNConv(MessagePassing):    def __init__(self, in_channels, out_channels):        super(GCNConv, self).__init__(aggr='add')  # "Add" aggregation.        self.lin = torch.nn.Linear(in_channels, out_channels)    def forward(self, x, edge_index):        # x has shape [N, in_channels]        # edge_index has shape [2, E]        # Step 1: 增加自连接到邻接矩阵        edge_index, _ = add_self_loops(edge_index, num_nodes=x.size(0))        # Step 2: 对节点的特征矩阵进行线性变换        x = self.lin(x)        # Step 3-5: Start propagating messages.        return self.propagate(edge_index, size=(x.size(0), x.size(0)), x=x)    def message(self, x_j, edge_index, size):        # x_j has shape [E, out_channels]        # Step 3: Normalize node features.        row, col = edge_index        deg = degree(row, size[0], dtype=x_j.dtype)        deg_inv_sqrt = deg.pow(-0.5)        norm = deg_inv_sqrt[row] * deg_inv_sqrt[col]        return norm.view(-1, 1) * x_j    def update(self, aggr_out):        # aggr_out has shape [N, out_channels]        # Step 5: Return new node embeddings.        return aggr_out