网络编程
位置:首页>> 网络编程>> Python编程>> pytorch geometric的GNN、GCN的节点分类方式

pytorch geometric的GNN、GCN的节点分类方式

作者:zhangztSky  发布时间:2022-12-24 16:01:23 

标签:pytorch,geometric,GNN,GCN

pytorch geometric的GNN、GCN节点分类

# -*- coding: utf-8 -*-

import os
import torch
import torch.nn.functional as F
import torch.nn as nn
from torch_geometric.datasets import Planetoid
import torch_geometric.nn as pyg_nn
import torch_geometric.transforms as T

# load dataset
def get_data(folder="node_classify/cora", data_name="cora"):
   # dataset = Planetoid(root=folder, name=data_name)
   dataset = Planetoid(root=folder, name=data_name,
                       transform=T.NormalizeFeatures())
   return dataset

# create the graph cnn model
class GraphCNN(nn.Module):
   def __init__(self, in_c, hid_c, out_c):
       super(GraphCNN, self).__init__()
       self.conv1 = pyg_nn.GCNConv(in_channels=in_c, out_channels=hid_c)
       self.conv2 = pyg_nn.GCNConv(in_channels=hid_c, out_channels=out_c)

def forward(self, data):
       # data.x data.edge_index
       x = data.x  # [N, C]
       edge_index = data.edge_index  # [2 ,E]

hid = self.conv1(x=x, edge_index=edge_index)  # [N, D]
       hid = F.relu(hid)

out = self.conv2(x=hid, edge_index=edge_index)  # [N, out_c]

out = F.log_softmax(out, dim=1)  # [N, out_c]

return out

class OwnGCN(nn.Module):
   def __init__(self, in_c, hid_c, out_c):
       super(OwnGCN, self).__init__()
       self.in_ = pyg_nn.SGConv(in_c, hid_c, K=2)

self.conv1 = pyg_nn.APPNP(K=2, alpha=0.1)
       self.conv2 = pyg_nn.APPNP(K=2, alpha=0.1)

self.out_ = pyg_nn.SGConv(hid_c, out_c, K=2)

def forward(self, data):
       x, edge_index = data.x, data.edge_index

x = self.in_(x, edge_index)
       x = F.dropout(x, p=0.1, training=self.training)

x = F.relu(self.conv1(x, edge_index))
       x = F.dropout(x, p=0.1, training=self.training)

x = F.relu(self.conv2(x, edge_index))
       x = F.dropout(x, p=0.1, training=self.training)

x = self.out_(x, edge_index)

return F.log_softmax(x, dim=1)

# todo list
class YourOwnGCN(nn.Module):
   pass

def analysis_data(dataset):
   print("Basic Info:      ", dataset[0])
   print("# Nodes:         ", dataset[0].num_nodes)
   print("# Features:      ", dataset[0].num_features)
   print("# Edges:         ", dataset[0].num_edges)
   print("# Classes:       ", dataset.num_classes)
   print("# Train samples: ", dataset[0].train_mask.sum().item())
   print("# Valid samples: ", dataset[0].val_mask.sum().item())
   print("# Test samples:  ", dataset[0].test_mask.sum().item())
   print("Undirected:      ", dataset[0].is_undirected())

def main():
   os.environ["CUDA_VISIBLE_DEVICES"] = "0"
   cora_dataset = get_data()

# todo list
   # my_net = GraphCNN(in_c=cora_dataset.num_features, hid_c=150, out_c=cora_dataset.num_classes)
   my_net = OwnGCN(in_c=cora_dataset.num_features, hid_c=300, out_c=cora_dataset.num_classes)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

my_net = my_net.to(device)
   data = cora_dataset[0].to(device)

optimizer = torch.optim.Adam(my_net.parameters(), lr=1e-2, weight_decay=1e-3)
   """
   # model train
   my_net.train()
   for epoch in range(500):
       optimizer.zero_grad()

output = my_net(data)
       loss = F.nll_loss(output[data.train_mask], data.y[data.train_mask])
       loss.backward()
       optimizer.step()

_, prediction = output.max(dim=1)

valid_correct = prediction[data.val_mask].eq(data.y[data.val_mask]).sum().item()
       valid_number = data.val_mask.sum().item()

valid_acc = valid_correct / valid_number
       print("Epoch: {:03d}".format(epoch + 1), "Loss: {:.04f}".format(loss.item()),
             "Valid Accuracy:: {:.4f}".format(valid_acc))
   """

# model test
   my_net = torch.load("node_classify/best.pth")
   my_net.eval()

_, prediction = my_net(data).max(dim=1)

target = data.y

test_correct = prediction[data.test_mask].eq(target[data.test_mask]).sum().item()
   test_number = data.test_mask.sum().item()

train_correct = prediction[data.train_mask].eq(target[data.train_mask]).sum().item()
   train_number = data.train_mask.sum().item()

print("==" * 20)

print("Accuracy of Train Samples: {:.04f}".format(train_correct / train_number))

print("Accuracy of Test  Samples: {:.04f}".format(test_correct / test_number))

def test_main():
   os.environ["CUDA_VISIBLE_DEVICES"] = "0"

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

cora_dataset = get_data()
   data = cora_dataset[0].to(device)

my_net = torch.load("node_classify/best.pth")

my_net.eval()
   _, prediction = my_net(data).max(dim=1)

target = data.y

test_correct = prediction[data.test_mask].eq(target[data.test_mask]).sum().item()
   test_number = data.test_mask.sum().item()

train_correct = prediction[data.train_mask].eq(target[data.train_mask]).sum().item()
   train_number = data.train_mask.sum().item()

print("==" * 20)

print("Accuracy of Train Samples: {:.04f}".format(train_correct / train_number))

print("Accuracy of Test  Samples: {:.04f}".format(test_correct / test_number))

if __name__ == '__main__':
   test_main()
   # main()
   # dataset = get_data()
   # analysis_data(dataset)

pytorch下GCN代码解读

def main():
   print("hello world")
main()

import os.path as osp
import argparse

import torch
import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
import torch_geometric.transforms as T
from torch_geometric.nn import GCNConv, ChebConv  # noqa

#GCN用于提取图的特征参数然后用于分类

#数据集初始化部分
parser = argparse.ArgumentParser()
parser.add_argument('--use_gdc', action='store_true',
                   help='Use GDC preprocessing.')
args = parser.parse_args()#是否使用GDC优化
dataset = 'CiteSeer'#训练用的数据集
path = osp.join(osp.dirname(osp.realpath(__file__)), '..', 'data', dataset)#数据集存放位置
dataset = Planetoid(path, dataset, transform=T.NormalizeFeatures())#数据初始化类,其dataset的基类是一个torch.utils.data.Dataset对象
data = dataset[0]#只有一个图作为训练数据
#print(data)

#预处理和模型定义
if args.use_gdc:
   gdc = T.GDC(self_loop_weight=1, normalization_in='sym',
               normalization_out='col',
               diffusion_kwargs=dict(method='ppr', alpha=0.05),
               sparsification_kwargs=dict(method='topk', k=128,
                                          dim=0), exact=True)
   data = gdc(data)#图扩散卷积用于预处理

#搭建模型
class Net(torch.nn.Module):
   #放置参数层(一般为可学习层,不可学习层也可放置,若不放置,则在forward中用functional实现)
   def __init__(self):
       super(Net, self).__init__()#在不覆盖Module的Init函数的情况下设置Net的init函数
       self.conv1 = GCNConv(dataset.num_features, 16, cached=True,
                            normalize=not args.use_gdc)#第一层GCN卷积运算输入特征向量大小为num_features输出大小为16
       #GCNConv的def init需要in_channel和out_channel(卷积核的数量)的参数,并对in_channel和out_channel调用linear函数,而该函数的作用为构建全连接层
       self.conv2 = GCNConv(16, dataset.num_classes, cached=True,
                            normalize=not args.use_gdc)#第二层GCN卷积运算输入为16(第一层的输出)输出为num_class
       # self.conv1 = ChebConv(data.num_features, 16, K=2)
       # self.conv2 = ChebConv(16, data.num_features, K=2)

#实现模型的功能各个层之间的连接关系(具体实现)
   def forward(self):
       x, edge_index, edge_weight = data.x, data.edge_index, data.edge_attr#赋值data.x特征向量edge_index图的形状,edge_attr权重矩阵
       x = F.relu(self.conv1(x, edge_index, edge_weight))#第一层用非线性激活函数relu
       #x,edge_index,edge_weight特征矩阵,邻接矩阵,权重矩阵组成GCN核心公式
       x = F.dropout(x, training=self.training)#用dropout函数防止过拟合
       x = self.conv2(x, edge_index, edge_weight)#第二层输出
       return F.log_softmax(x, dim=1)#log_softmax激活函数用于最后一层返回分类结果
#Z=log_softmax(A*RELU(A*X*W0)*W1)A连接关系X特征矩阵W参数矩阵
#得到Z后即可用于分类
#softmax(dim=1)行和为1再取log  x为节点的embedding

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')#指定设备
model, data = Net().to(device), data.to(device)#copy model,data到device上

#优化算法
optimizer = torch.optim.Adam([
   dict(params=model.conv1.parameters(), weight_decay=5e-4),#权重衰减避免过拟合
   dict(params=model.conv2.parameters(), weight_decay=0)#需要优化的参数
], lr=0.01)  # Only perform weight-decay on first convolution.
#lr步长因子或者是学习率

#模型训练
def train():
   model.train()#设置成train模式
   optimizer.zero_grad()#清空所有被优化的变量的梯度
   F.nll_loss(model()[data.train_mask], data.y[data.train_mask]).backward()#损失函数训练参数用于节点分类
   optimizer.step()#步长

@torch.no_grad()#不需要计算梯度,也不进行反向传播

#测试
def test():
   model.eval()#设置成evaluation模式
   logits, accs = model(), []
   for _, mask in data('train_mask', 'val_mask', 'test_mask'):#mask矩阵,掩膜作用与之相应的部分不会被更新
       pred = logits[mask].max(1)[1]#mask对应点的输出向量最大值并取序号1
       acc = pred.eq(data.y[mask]).sum().item() / mask.sum().item()#判断是否相等计算准确度
       accs.append(acc)
   return accs

best_val_acc = test_acc = 0

#执行
for epoch in range(1, 201):
   train()
   train_acc, val_acc, tmp_test_acc = test()#训练准确率,实际输入的准确率,测试准确率
   if val_acc > best_val_acc:
       best_val_acc = val_acc
       test_acc = tmp_test_acc
   log = 'Epoch: {:03d}, Train: {:.4f}, Val: {:.4f}, Test: {:.4f}'#类型及保留位数
   print(log.format(epoch, train_acc, best_val_acc, test_acc))#输出格式化函数'''

来源:https://blog.csdn.net/qq_38574975/article/details/107443725

0
投稿

猜你喜欢

手机版 网络编程 asp之家 www.aspxhome.com