数据完整存储与内存的数据集类
使用数据集的一般过程
PyG使用数据的一般过程:
- 从网络上下载数据原始文件;
- 对原始数据文件做处理,为每一个图样本生成一个
Data
对象; - 对每一个
Data
对象执行数据处理,使其转换成新的Data
对象; - 过滤
Data
对象; - 保存
Data
对象到文件; - 获取
Data
对象,在每一次获取Data
对象时,都先对Data
对象做数据变换,所以获取到的是数据变换后的Data
对象)。
实际中并非需要严格执行每一个步骤
占用内存较小的数据集,我们可以将整个数据集的数据都存储到内存里。PyG提供数据完全存于内存的数据集类(InMemoryDataset
)。
class InMemoryDataset(root: Optional[str] = None, transform: Optional[Callable] = None, pre_transform: Optional[Callable] = None, pre_filter: Optional[Callable] = None)
InMemoryDataset
类初始化方法参数说明:
root
:存储数据集的文件夹的路径。该文件夹下有两个文件夹:raw_dir
,用于存储未处理的文件,从网络下载的数据集原始文件会被存放到这里;processed_dir
,处理后的数据被保存到这里,从此文件夹下加载文件即可获得Data
对象。
transform
:函数类型,一个数据转换函数,接收Data
对象并返回转换后的Data
对象。此函数在每一次数据获取过程中都会被执行。获取数据的函数首先使用此函数对Data
对象做转换,然后才返回数据。此函数应该用于数据增广(Data Augmentation)。默认为None
,不对数据做转换。pre_transform
:函数类型,一个数据转换函数,它接收Data
对象并返回转换后的Data
对象。此函数在Data
对象被保存到文件前调用。因此它应该用于只执行一次的数据预处理。该参数默认值为None
,表示不做数据预处理。pre_filter
:函数类型,检查数据是否要保留的函数,它接收Data
对象,返回此Data
对象是否应该被包含在最终的数据集中。此函数也在[Data
]对象被保存到文件前调用。默认为None
,表示不做数据检查,保留所有的数据。
通过继承[InMemoryDataset
]类来构造一个我们自己的数据集类,我们需要实现四个基本方法:
raw_file_names()
:属性方法,返回一个数据集原始文件的文件名列表,若在raw_dir
文件夹中找不到数据集原始文件,则调用process()
函数下载文件到raw_dir
文件夹。processed_file_names()
。属性方法,返回一个存储处理过的数据的文件的文件名列表,若在processed_dir
文件夹中找不到存储处理过的数据的文件,则调用process()
函数对样本做处理,然后保存处理过的数据到processed_dir
文件夹下的文件里。download()
: 下载数据集原始文件到raw_dir
文件夹。process()
: 处理数据,保存处理好的数据到processed_dir
文件夹下的文件。
通过继承自InMemoryDataset
自定义的数据集类
import torch
from torch_geometric.data import InMemoryDataset, download_url
class MyOwnDataset(InMemoryDataset):
def __init__(self, root, transform=None, pre_transform=None, pre_filter=None):
super().__init__(root=root, transform=transform, pre_transform=pre_transform, pre_filter=pre_filter)
self.data, self.slices = torch.load(self.processed_paths[0])
@property
def raw_file_names(self):
return ['some_file_1', 'some_file_2', ...]
@property
def processed_file_names(self):
return ['data.pt']
def download(self):
# Download to `self.raw_dir`.
download_url(url, self.raw_dir)
...
def process(self):
# Read data into huge `Data` list.
data_list = [...]
if self.pre_filter is not None:
data_list = [data for data in data_list if self.pre_filter(data)]
if self.pre_transform is not None:
data_list = [self.pre_transform(data) for data in data_list]
data, slices = self.collate(data_list)
torch.save((data, slices), self.processed_paths[0])
InMemoryDataset
数据集类实例
PubMed
数据集存储的是文章引用网络,文章对应图的结点,如果两篇文章存在引用关系,则这两篇文章对应的结点之间存在边。PyG中的Planetoid
数据集类包含了数据集PubMed
的使用,因此我们直接基于Planetoid
类进行修改,得到PlanetoidPubMed
数据集类。
PubMed
数据集来源于论文Revisiting Semi-Supervised Learning with Graph Embeddings。
构造PlanetoidPubMed
数据集类
import os.path as osp
import torch
from torch_geometric.data import (InMemoryDataset, download_url)
from torch_geometric.io import read_planetoid_data
class PlanetoidPubMed(InMemoryDataset):
url = 'https://github.com/kimiyoung/planetoid/raw/master/data'
def __init__(self, root, transform=None, pre_transform=None):
super(PlanetoidPubMed, self).__init__(root, transform, pre_transform)
self.data, self.slices = torch.load(self.processed_paths[0])
@property
def raw_dir(self):
return osp.join(self.root, 'raw')
@property
def processed_dir(self):
return osp.join(self.root, 'processed')
@property
def raw_file_names(self):
names = ['x', 'tx', 'allx', 'y', 'ty', 'ally', 'graph', 'test.index']
return ['ind.pubmed.{}'.format(name) for name in names]
@property
def processed_file_names(self):
return 'data.pt'
def download(self):
for name in self.raw_file_names:
download_url('{}/{}'.format(self.url, name), self.raw_dir)
def process(self):
data = read_planetoid_data(self.raw_dir, 'pubmed')
data = data if self.pre_transform is None else self.pre_transform(data)
torch.save(self.collate([data]), self.processed_paths[0])
def __repr__(self):
return '{}()'.format(self.name)
查看这个数据集
dataset = PlanetoidPubMed('dataset/PlanetoidPubMed')
print(dataset.num_classes)
print(dataset[0].num_nodes)
print(dataset[0].num_edges)
print(dataset[0].num_features)
Processing...
Done!
3
19717
88648
500
节点预测与边预测任务实践
1.节点预测
之前设计过2层GATConv
组成的图神经网络,现重定义一个GAT网络,使其能够通过参数来定义GATConv
的层数,以及每一层GATConv
的out_channels
。
import os.path as osp
import torch
from torch_geometric.data import (InMemoryDataset, download_url)
from torch_geometric.io import read_planetoid_data
from torch_geometric.transforms import NormalizeFeatures
from torch_geometric.nn import Sequential, GATConv
from torch.nn import ReLU, Linear
import torch.nn.functional as F
class GAT(torch.nn.Module):
def __init__(self, num_features, hidden_channels_list, num_classes):
super(GAT, self).__init__()
torch.manual_seed(12345)
hns = [num_features] + hidden_channels_list
conv_list = []
for idx in range(len(hidden_channels_list)):
conv_list.append((GATConv(hns[idx], hns[idx+1]), 'x, edge_index -> x'))
conv_list.append(ReLU(inplace=True),)
self.convseq = Sequential('x, edge_index', conv_list)
self.linear = Linear(hidden_channels_list[-1], num_classes)
def forward(self, x, edge_index):
x = self.convseq(x, edge_index)
x = F.dropout(x, p=0.5, training=self.training)
x = self.linear(x)
return x
定义网络训练和测试
def train(model, data, criterion, optimizer):
model.train()
optimizer.zero_grad() # Clear gradients.
out = model(data.x, data.edge_index) # Perform a single forward pass.
loss = criterion(out[data.train_mask],
data.y[data.train_mask]) # Compute the loss solely based on the training nodes.
loss.backward() # Derive gradients.
optimizer.step() # Update parameters based on gradients.
return loss
def test(model, data):
model.eval()
out = model(data.x, data.edge_index)
pred = out.argmax(dim=1) # Use the class with highest probability.
test_correct = pred[data.test_mask] == data.y[data.test_mask] # Check against ground-truth labels.
test_acc = int(test_correct.sum()) / int(data.test_mask.sum()) # Derive ratio of correct predictions.
return test_acc
我们来训练一下这个网络看看效果
def run_gat():
dataset = PlanetoidPubMed(root='dataset/PlanetoidPubMed/', transform=NormalizeFeatures())
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
data = dataset[0].to(device)
hidden_channels_list = [100, 200]
model = GAT(num_features=dataset.num_features,
hidden_channels_list=hidden_channels_list,
num_classes=dataset.num_classes).to(device)
print(model)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
criterion = torch.nn.CrossEntropyLoss()
# init_lr = optimizer.param_groups[0]['lr']
for epoch in range(1, 501):
# adjust_learning_rate(optimizer, epoch, init_lr)
# lr = optimizer.param_groups[0]['lr']
# print(epoch, lr)
loss = train(model, data, criterion, optimizer)
print(f'Epoch: {epoch:03d}, Loss: {loss:.4f}')
test_accuray = test(model, data)
print(f'Test Accuracy: {test_accuray:.4f}')
GAT(
(convseq): Sequential(
(0): GATConv(500, 100, heads=1)
(1): ReLU(inplace=True)
(2): GATConv(100, 200, heads=1)
(3): ReLU(inplace=True)
)
(linear): Linear(in_features=200, out_features=3, bias=True)
)
Epoch: 001, Loss: 1.0990
Epoch: 002, Loss: 1.0965
Epoch: 003, Loss: 1.0926
Epoch: 004, Loss: 1.0850
Epoch: 005, Loss: 1.0739
Epoch: 006, Loss: 1.0524
Epoch: 007, Loss: 1.0260
Epoch: 008, Loss: 0.9887
Epoch: 009, Loss: 0.9284
Epoch: 010, Loss: 0.8804
...
Epoch: 496, Loss: 0.0060
Epoch: 497, Loss: 0.0069
Epoch: 498, Loss: 0.0064
Epoch: 499, Loss: 0.0076
Epoch: 500, Loss: 0.0060
Test Accuracy: 0.7630
2.边预测
边预测任务,目标是预测两个节点之间是否存在边。
对于一个图数据集,我们有节点属性x
,边端点edge_index
。edge_index
存储的就是正样本。为了实现边预测任务,我们需要生成部分负样本,即采样一些不存在边的节点对作为负样本边,正负样本数量应相当, 以便保持平衡。数据集拆分为训练集、验证集和测试集三个集合。
PyG提供了采样负样本边的方法
train_test_split_edges(data, val_ratio=0.05, test_ratio=0.1)
- data参数为
torch_geometric.data.Data
对象, - val_ratio参数为验证集所占比例,
- test_ratio参数为测试集所占比例。
该函数将自动采样得到负样本,并将正负样本拆分成训练集、验证集和测试集三个集合。edge_index
属性转为train_pos_edge_index
、train_neg_adj_mask
、val_pos_edge_index
、val_neg_edge_index
、test_pos_edge_index
和test_neg_edge_index
六个属性。
注:train_neg_adj_mask
与其他属性格式不同,其实该属性在后面并没有派上用场,后面我们仍然需要进行一次训练集负样本采样。
下面我们使用Cora
数据集作为例子,进行边预测任务说明。
from torch_geometric.datasets import Planetoid
import torch_geometric.transforms as T
from torch_geometric.utils import train_test_split_edges
# if __name__ == '__main__':
dataset = Planetoid('dataset', 'Cora', transform=T.NormalizeFeatures())
data = dataset[0]
data.train_mask = data.val_mask = data.test_mask = data.y = None
print(data.edge_index.shape)
data = train_test_split_edges(data)
for key in data.keys:
print(key, getattr(data, key).shape)
torch.Size([2, 10556])
x torch.Size([2708, 1433])
val_pos_edge_index torch.Size([2, 263])
test_pos_edge_index torch.Size([2, 527])
train_pos_edge_index torch.Size([2, 8976])
train_neg_adj_mask torch.Size([2708, 2708])
val_neg_edge_index torch.Size([2, 263])
test_neg_edge_index torch.Size([2, 527])
训练集、验证集和测试集中正样本边的数量之和不等于原始边的数量。因为Cora
图是无向图,在统计原始边数量时,统计每一条边正向与反向各计了一次,训练集也包含边的正向与反向,但验证集与测试集都只包含了边的一个方向。因为训练集用于训练时一条边的两个端点要互传信息,只考虑一个方向的话,表示只能由一个端点传信息给另一个端点,而验证集与测试集的边用于衡量检验边预测的准确性,只需考虑一个方向的边即可。
边预测图神经网络的构造
import torch
from torch_geometric.nn import GCNConv
class Net(torch.nn.Module):
def __init__(self, in_channels, out_channels):
super(Net, self).__init__()
self.conv1 = GCNConv(in_channels, 128)
self.conv2 = GCNConv(128, out_channels)
def encode(self, x, edge_index):
x = self.conv1(x, edge_index)
x = x.relu()
return self.conv2(x, edge_index)
def decode(self, z, pos_edge_index, neg_edge_index):
edge_index = torch.cat([pos_edge_index, neg_edge_index], dim=-1)
return (z[edge_index[0]] * z[edge_index[1]]).sum(dim=-1)
def decode_all(self, z):
prob_adj = z @ z.t()
return (prob_adj > 0).nonzero(as_tuple=False).t()
定义单个epoch的训练过程
def get_link_labels(pos_edge_index, neg_edge_index):
num_links = pos_edge_index.size(1) + neg_edge_index.size(1)
link_labels = torch.zeros(num_links, dtype=torch.float)
link_labels[:pos_edge_index.size(1)] = 1.
return link_labels
def train(data, model, optimizer):
model.train()
neg_edge_index = negative_sampling(
edge_index=data.train_pos_edge_index,
num_nodes=data.num_nodes,
num_neg_samples=data.train_pos_edge_index.size(1))
optimizer.zero_grad()
z = model.encode(data.x, data.train_pos_edge_index)
link_logits = model.decode(z, data.train_pos_edge_index, neg_edge_index)
link_labels = get_link_labels(data.train_pos_edge_index, neg_edge_index).to(data.x.device)
loss = F.binary_cross_entropy_with_logits(link_logits, link_labels)
loss.backward()
optimizer.step()
return loss
定义测试方法
@torch.no_grad()
def test(data, model):
model.eval()
z = model.encode(data.x, data.train_pos_edge_index)
results = []
for prefix in ['val', 'test']:
pos_edge_index = data[f'{prefix}_pos_edge_index']
neg_edge_index = data[f'{prefix}_neg_edge_index']
link_logits = model.decode(z, pos_edge_index, neg_edge_index)
link_probs = link_logits.sigmoid()
link_labels = get_link_labels(pos_edge_index, neg_edge_index)
results.append(roc_auc_score(link_labels.cpu(), link_probs.cpu()))
return results
运行完整的训练、验证与测试
def main():
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
dataset = 'Cora'
dataset = Planetoid('dataset', dataset, transform=T.NormalizeFeatures())
data = dataset[0]
ground_truth_edge_index = data.edge_index.to(device)
data.train_mask = data.val_mask = data.test_mask = data.y = None
data = train_test_split_edges(data)
data = data.to(device)
model = Net(dataset.num_features, 64).to(device)
optimizer = torch.optim.Adam(params=model.parameters(), lr=0.01)
best_val_auc = test_auc = 0
for epoch in range(1, 301):
loss = train(data, model, optimizer)
val_auc, tmp_test_auc = test(data, model)
if val_auc > best_val_auc:
best_val_auc = val_auc
test_auc = tmp_test_auc
print(f'Epoch: {epoch:03d}, Loss: {loss:.4f}, Val Accuracy: {val_auc:.4f}, '
f'Test Accuracy: {test_auc:.4f}')
z = model.encode(data.x, data.train_pos_edge_index)
final_edge_index = model.decode_all(z)
训练结果
Epoch: 001, Loss: 0.6930, Val Accuracy: 0.6852, Test Accuracy: 0.6838
Epoch: 002, Loss: 0.6816, Val Accuracy: 0.6784, Test Accuracy: 0.6838
Epoch: 003, Loss: 0.7089, Val Accuracy: 0.6838, Test Accuracy: 0.6838
Epoch: 004, Loss: 0.6767, Val Accuracy: 0.6979, Test Accuracy: 0.6983
Epoch: 005, Loss: 0.6843, Val Accuracy: 0.7087, Test Accuracy: 0.7065
...
Epoch: 096, Loss: 0.4456, Val Accuracy: 0.9141, Test Accuracy: 0.9084
Epoch: 097, Loss: 0.4450, Val Accuracy: 0.9146, Test Accuracy: 0.9084
Epoch: 098, Loss: 0.4357, Val Accuracy: 0.9148, Test Accuracy: 0.9084
Epoch: 099, Loss: 0.4415, Val Accuracy: 0.9137, Test Accuracy: 0.9084
Epoch: 100, Loss: 0.4431, Val Accuracy: 0.9129, Test Accuracy: 0.9084