当前位置: 首页 > news >正文

关于 ogbg-molhi数据集的个人解析

 cs224w_colab2.py这个图属性预测到底咋预测的

dataset.meta_info.T
Out[2]:
num tasks                                                                1 
eval metric                                                         rocauc
download_name                                                          hiv
version                                                                  1
url                      http://snap.stanford.edu/ogb/data/graphproppre...
add_inverse_edge                                                      True
data type                                                              mol
has_node_attr                                                         True
has_edge_attr                                                         True
task type                                            binary classification
num classes                                                              2
split                                                             scaffold
additional node files                                                 None
additional edge files                                                 None
binary                                                               False
Name: ogbg-molhiv, dtype: object

参照上面 这里的num tasks  仅适用于图属性预测? num tasks = 1

model = GCN_Graph(args['hidden_dim'],dataset.num_tasks, args['num_layers'],args['dropout']).to(device)train_loader.dataset.data.edge_index.shape
Out[10]: torch.Size([2, 2259376])train_loader.dataset.data.edge_attr.shape
Out[12]: torch.Size([2259376, 3])type(train_loader.dataset.data.node_stores)
Out[26]: listtrain_loader.dataset.data.node_stores[0]['y'].shape
Out[46]: torch.Size([41127, 1])
train_loader.dataset.data.node_stores[0]['y'].sum()
Out[47]: tensor(1443) y 中的数值求和值torch.unique(train_loader.dataset.data.node_stores[0]['y'],return_counts=True)
Out[58]: (tensor([0, 1]), tensor([39684,  1443]))  仅0,1两类self.node_encoder.atom_embedding_list
Out[62]: 
ModuleList((0): Embedding(119, 256)(1): Embedding(5, 256)(2): Embedding(12, 256)(3): Embedding(12, 256)(4): Embedding(10, 256)(5): Embedding(6, 256)(6): Embedding(6, 256)(7): Embedding(2, 256)(8): Embedding(2, 256)
)list(enumerate(data_loader))
Out[82]: 
[(0,DataBatch(edge_index=[2, 1734], edge_attr=[1734, 3], x=[807, 9], y=[32, 1], num_nodes=807, batch=[807], ptr=[33])),若干组 很多
        x, edge_index, batch = batched_data.x, batched_data.edge_index, batched_data.batchembed = self.node_encoder(x)  #使用编码器 将原先9维的编码为256维 self.node_encoder = AtomEncoder(hidden_dim)out = self.gnn_node(embed, edge_index) #使用gcn得到节点嵌入 embed=X edge_index 连边/节点对 out = self.pool(out, batch)batch.unique(return_counts = True)
Out[94]: 
(tensor([ 0,  1,  2,  3,  4,  5,  6,  7,  8,  9, 10, 11, 12, 13, 14, 15, 16, 17,18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31],device='cuda:0'), 这里说明有31个待训练子图(化学分子) 下图api说明了聚合过程tensor([30, 18, 21, 26, 12, 20, 17, 18, 36, 11, 31, 22, 21, 26, 22, 21, 21, 63,15, 18, 18, 29, 18, 40, 41, 19, 19, 30, 12, 21, 19, 23], 每个分子中所含有的节点(原子)数量device='cuda:0'))
batch.shape
Out[95]: torch.Size([758])

def global_mean_pool(x: Tensor, batch: Optional[Tensor],size: Optional[int] = None) -> Tensor:dim = -1 if x.dim() == 1 else -2 #这里的x.dim() = 2 
# dim() → int Returns the number of dimensions of self tensor.if batch is None:return x.mean(dim=dim, keepdim=x.dim() <= 2) #keepdim=x.dim() <= 2 ??啥玩意<=size = int(batch.max().item() + 1) if size is None else sizereturn scatter(x, batch, dim=dim, dim_size=size, reduce='mean')

This package consists of a small extension library of highly optimized sparse update (scatter and segment) operations for the use in PyTorch, which are missing in the main package. Scatter and segment operations can be roughly described as reduce operations based on a given "group-index" tensor. Segment operations require the "group-index" tensor to be sorted, whereas scatter operations are not subject to these requirements.该包由一个小型扩展库组成,该库包含用于PyTorch的高度优化的稀疏更新(分散和分段)操作,这些操作在主包中丢失。散射和分段运算可以粗略地描述为基于给定“群索引”张量的归约运算。分段运算需要对“组索引”张量进行排序,而分散运算则不受这些要求的约束。

由此(scatter)由多个节点的嵌入值最终得到这部分节点所在的子图嵌入(化学分子)。

    def forward(self, batched_data):# TODO: Implement a function that takes as input a# mini-batch of graphs (torch_geometric.data.Batch) and# returns the predicted graph property for each graph.## NOTE: Since we are predicting graph level properties,# your output will be a tensor with dimension equaling# the number of graphs in the mini-batchx, edge_index, batch = batched_data.x, batched_data.edge_index, batched_data.batchembed = self.node_encoder(x)  #使用编码器 将原先9维的编码为256维 self.node_encoder = AtomEncoder(hidden_dim)out = self.gnn_node(embed, edge_index) #使用gcn得到节点嵌入 embed=X edge_index 连边/节点对out = self.pool(out, batch)out = self.linear(out)############# Your code here ############## Note:## 1. Construct node embeddings using existing GCN model## 2. Use the global pooling layer to aggregate features for each individual graph## For more information please refer to the documentation:## https://pytorch-geometric.readthedocs.io/en/latest/modules/nn.html#global-pooling-layers## 3. Use a linear layer to predict each graph's property## (~3 lines of code)#########################################return outout.shape
Out[122]: torch.Size([32, 1])
out
Out[121]: 
tensor([[-0.4690],[-1.0285],[-0.4614],
最后经过线性层 返回得到所属类别概率 运行到如下部分结束反向传播forward() (op = model(batch)# 先进入model函数 然后运行 反向传播)def train(model, device, data_loader, optimizer, loss_fn):# TODO: Implement a function that trains your model by# using the given optimizer and loss_fn.model.train()  #Sets the module in training mode. data_loader.dataset.data Data(num_nodes=1049163, edge_index=[2, 2259376], edge_attr=[2259376, 3], x=[1049163, 9], y=[41127, 1])loss = 0for step, batch in enumerate(tqdm(data_loader, desc="Iteration")): #,total= data_loader.batch_sampler# for step, batch in tqdm(enumerate(data_loader), desc="Iteration"): #,total= data_loader.batch_samplerbatch = batch.to(device)if batch.x.shape[0] == 1 or batch.batch[-1] == 0:passelse:## ignore nan targets (unlabeled) when computing training loss.is_labeled = batch.y == batch.y # 0/1转化为Ture/False############# Your code here ############## Note:## 1. Zero grad the optimizer## 2. Feed the data into the model## 3. Use `is_labeled` mask to filter output and labels## 4. You may need to change the type of label to torch.float32## 5. Feed the output and label to the loss_fn## (~3 lines of code)optimizer.zero_grad()# print('optimizer.zero_grad()')op = model(batch)# 先进入model函数 然后运行 反向传播。。。。。。。。。。。。。。。后面计算损失 更新梯度等等

存在错误等欢迎指正! 附件为整个作业的.py文件

http://www.lryc.cn/news/162832.html

相关文章:

  • RabbitMQ:hello结构
  • SpringBoot整合Redis 并 展示使用方法
  • js中如何实现字符串去重?
  • Axure RP仿QQ音乐app高保真原型图交互模板源文件
  • 2023牛客暑假多校第四场(补题向题解:J)
  • 第 362 场 LeetCode 周赛题解
  • C++ if 语句
  • 业务安全及实战案例
  • 十一)Stable Diffussion使用教程:人物三视图
  • 超级等级福利礼包
  • 如何用Jmeter提取和引用Token
  • C#文件拷贝工具
  • Redis——Java中的客户端和API
  • Brief. Bioinformatics2021 | sAMP-PFPDeep+:利用三种不同的序列编码和深度神经网络预测短抗菌肽
  • 问道管理:华为产业链股再度拉升,捷荣技术6连板,华力创通3日大涨近70%
  • 面试设计模式-责任链模式
  • Qt 开发 CMake工程
  • 2.k8s账号密码登录设置
  • 【代表团坐车】Python 实现-附ChatGPT解析
  • 【Java】x-easypdf: 一种简单易用的PDF处理库
  • 1 Linux输入子系统
  • Zabbix 利用 Grafana 进行图形展示
  • 【LeetCode周赛】LeetCode第362场周赛
  • Leetcode128. 最长连续序列
  • K8S:kubeadm搭建K8S+Harbor 私有仓库
  • MaskVO: Self-Supervised Visual Odometry with a Learnable Dynamic Mask 论文阅读
  • 面试求职-面试注意事项
  • sm2 签名验签
  • 如何检查Windows 11笔记本电脑电池健康状况
  • 编程大师-分布式