GCN实战深入浅出图神经网络第五章:基于Cora数据集的GCN节点分类 代码分析
SetUp,库声明
In [2]:
import itertools
import os
import os.path as osp
import pickle
import urllib
from collections import namedtuple
import numpy as np
import scipy.sparse as sp
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.nn.init as init
import torch.optim as optim
import matplotlib.pyplot as plt
%matplotlib inline
数据准备
Cora数据集说明
├── gcn
│ ├── data //图数据
│ │ ├── ind.citeseer.allx
│ │ ├── ind.citeseer.ally
│ │ ├── ind.citeseer.graph
│ │ ├── ind.citeseer.test.index
│ │ ├── ind.citeseer.tx # 1
│ │ ├── ind.citeseer.ty # 2
│ │ ├── ind.citeseer.x # 3
│ │ ├── ind.citeseer.y # 4
│ │ ├── ind.cora.allx
│ │ ├── ind.cora.ally
│ │ ├── ind.cora.graph
│ │ ├── ind.cora.test.index
│ │ ├── ind.cora.tx
│ │ ├── ind.cora.ty
│ │ ├── ind.cora.x
│ │ ├── ind.cora.y
│ │ ├── ind.pubmed.allx
│ │ ├── ind.pubmed.ally
│ │ ├── ind.pubmed.graph
│ │ ├── ind.pubmed.test.index
│ │ ├── ind.pubmed.tx
│ │ ├── ind.pubmed.ty
│ │ ├── ind.pubmed.x
│ │ └── ind.pubmed.y
│ ├── __init__.py
│ ├── inits.py //初始化的公用函数
│ ├── layers.py //GCN层定义
│ ├── metrics.py //评测指标的计算
│ ├── models.py //模型结构定义
│ ├── train.py //训练
│ └── utils.py //工具函数的定义
├── LICENCE
├── README.md
├── requirements.txt
└── setup.py
实验时可能出现数据集下载出错的问题,可以自行下载,放进程序预定的文件夹即可
链接:https://pan.baidu.com/s/13JZDlHAlPow0FRtkwzJN8Q
提取码:7fay
cora读取的文件说明:
ind.cora.x => 训练实例的特征向量,是scipy.sparse.csr.csr_matrix类对象,shape:(140, 1433),由于ind.cora.x的数据包含于 allx 中,所以实验中没有读取 x
ind.cora.tx => 测试实例的特征向量,shape:(1000, 1433)
ind.cora.allx => 有标签的+无无标签训练实例的特征向量,是ind.dataset_str.x的超集,shape:(1708, 1433)
# 实验中的完整的特征向量是有(allx,tx)拼接而成,(2708,1433),在实际训练是整体训练,只有当要计算损失值和精确度时才用掩码从(allx,tx)截取相应的输出
ind.cora.y => 训练实例的 标签,独热编码,numpy.ndarray类的实例,是numpy.ndarray对象,shape:(140, 7)
ind.cora.ty => 测试实例的标签,独热编码,numpy.ndarray类的实例,shape:(1000, 7)
ind.cora.ally => 对应于ind.dataset_str.allx的标签,独热编码,shape:(1708, 7)
# 同样是(ally,ty)拼接
ind.cora.graph => 图数据,collections.defaultdict类的实例,格式为 {index:[index_of_neighbor_nodes]}
ind.cora.test.index => 测试实例的id,(1000,)
In [3]:
Data = namedtuple('Data', ['x', 'y', 'adjacency',
'train_mask', 'val_mask', 'test_mask'])
def tensor_from_numpy(x, device):
return torch.from_numpy(x).to(device)
class CoraData(object):
download_url = "https://github.com/kimiyoung/planetoid/raw/master/data"
filenames = ["ind.cora.{}".format(name) for name in
['x', 'tx', 'allx', 'y', 'ty', 'ally', 'graph', 'test.index']]
def __init__(self, data_root="cora", rebuild=False):
"""Cora数据,包括数据下载,处理,加载等功能
当数据的缓存文件存在时,将使用缓存文件,否则将下载、进行处理,并缓存到磁盘
处理之后的数据可以通过属性 .data 获得,它将返回一个数据对象,包括如下几部分:
* x: 节点的特征,维度为 2708 * 1433,类型为 np.ndarray
* y: 节点的标签,总共包括7个类别,类型为 np.ndarray
* adjacency: 邻接矩阵,维度为 2708 * 2708,类型为 scipy.sparse.coo.coo_matrix
* train_mask: 训练集掩码向量,维度为 2708,当节点属于训练集时,相应位置为True,否则False
* val_mask: 验证集掩码向量,维度为 2708,当节点属于验证集时,相应位置为True,否则False
* test_mask: 测试集掩码向量,维度为 2708,当节点属于测试集时,相应位置为True,否则False
Args:
-------
data_root: string, optional
存放数据的目录,原始数据路径: {data_root}/raw
缓存数据路径: {data_root}/processed_cora.pkl
rebuild: boolean, optional
是否需要重新构建数据集,当设为True时,如果存在缓存数据也会重建数据
"""
self.data_root = data_root
save_file = osp.join(self.data_root, "processed_cora.pkl")
if osp.exists(save_file) and not rebuild:
print("Using Cached file: {}".format(save_file))
self._data = pickle.load(open(save_file, "rb"))
else:
self.maybe_download()
self._data = self.process_data()
with open(save_file, "wb") as f:
pickle.dump(self.data, f)
print("Cached file: {}".format(save_file))
@property
def data(self):
"""返回Data数据对象,包括x, y, adjacency, train_mask, val_mask, test_mask"""
return self._data
def process_data(self):
"""
处理数据,得到节点特征和标签,邻接矩阵,训练集、验证集以及测试集
引用自:https://github.com/rusty1s/pytorch_geometric
"""
print("Process data ...")
_, tx, allx, y, ty, ally, graph, test_index = [self.read_data(
osp.join(self.data_root, "raw", name)) for name in self.filenames]
# 测试test_index的形状(1000,),如果那里不明白可以测试输出一下矩阵形状
print('test_index',test_index.shape)
train_index = np.arange(y.shape[0]) # [0,...139] 140个元素
print('train_index',train_index.shape)
val_index = np.arange(y.shape[0], y.shape[0] + 500) # [140 - 640] 500个元素
print('val_index',val_index.shape)
sorted_test_index = sorted(test_index) # #test_index就是随机选取的下标,排下序
# print('test_index',sorted_test_index)
x = np.concatenate((allx, tx), axis=0) # 1708 +1000 =2708 特征向量
y = np.concatenate((ally, ty), axis=0).argmax(axis=1) # 把最大值的下标重新作为一个数组 标签向量
x[test_index] = x[sorted_test_index] # 打乱顺序,单纯给test_index 的数据排个序,不清楚具体效果
y[test_index] = y[sorted_test_index]
num_nodes = x.shape[0] #2078
train_mask = np.zeros(num_nodes, dtype=np.bool) # 生成零向量
val_mask = np.zeros(num_nodes, dtype=np.bool)
test_mask = np.zeros(num_nodes, dtype=np.bool)
train_mask[train_index] = True # 前140个元素为训练集
val_mask[val_index] = True # 140 -639 500个
test_mask[test_index] = True # 1708-2708 1000个元素
#下面两句是我测试用的,本来代码没有
#是为了知道使用掩码后,y_train_mask 的形状,由输出来说是(140,)
# 这就是后面划分训练集的方法
y_train_mask = y[train_mask]
print('y_train_mask',y_train_mask.shape)
#构建邻接矩阵
adjacency = self.build_adjacency(graph)
print("Node's feature shape: ", x.shape)
print("Node's label shape: ", y.shape)
print("Adjacency's shape: ", adjacency.shape)
print("Number of training nodes: ", train_mask.sum())
print("Number of validation nodes: ", val_mask.sum())
print("Number of test nodes: ", test_mask.sum())
return Data(x=x, y=y, adjacency=adjacency,
train_mask=train_mask, val_mask=val_mask, test_mask=test_mask)
def maybe_download(self):
save_path = os.path.join(self.data_root, "raw")
for name in self.filenames:
if not osp.exists(osp.join(save_path, name)):
self.download_data(
"{}/{}".format(self.download_url, name), save_path)
@staticmethod
def build_adjacency(adj_dict):
"""根据邻接表创建邻接矩阵"""
edge_index = []
num_nodes = len(adj_dict)
print('num_nodesaaaaaaaaaaaa',num_nodes)
for src, dst in adj_dict.items(): # 格式为 {index:[index_of_neighbor_nodes]}
edge_index.extend([src, v] for v in dst) #
edge_index.extend([v, src] for v in dst)
# 去除重复的边
edge_index = list(k for k, _ in itertools.groupby(sorted(edge_index))) # 以轮到的元素为k,每个k对应一个数组,和k相同放进数组,不
# 同再生成一个k,sorted()是以第一个元素大小排序
edge_index = np.asarray(edge_index)
#稀疏矩阵 存储非0值 节省空间
adjacency = sp.coo_matrix((np.ones(len(edge_index)),
(edge_index[:, 0], edge_index[:, 1])),
shape=(num_nodes, num_nodes), dtype="float32")
return adjacency
@staticmethod
def read_data(path):
"""使用不同的方式读取原始数据以进一步处理"""
name = osp.basename(path)
if name == "ind.cora.test.index":
out = np.genfromtxt(path, dtype="int64")
return out
else:
out = pickle.load(open(path, "rb"), encoding="latin1")
out = out.toarray() if hasattr(out, "toarray") else out
return out
@staticmethod
def download_data(url, save_path):
"""数据下载工具,当原始数据不存在时将会进行下载"""
if not os.path.exists(save_path):
os.makedirs(save_path)
data = urllib.request.urlopen(url)
filename = os.path.split(url)[-1]
with open(os.path.join(save_path, filename), 'wb') as f:
f.write(data.read())
return True
@staticmethod
def normalization(adjacency):
"""计算 L=D^-0.5 * (A+I) * D^-0.5"""
adjacency += sp.eye(adjacency.shape[0]) # 增加自连接
degree = np.array(adjacency.sum(1))
d_hat = sp.diags(np.power(degree, -0.5).flatten())
return d_hat.dot(adjacency).dot(d_hat).tocoo() #返回稀疏矩阵的coo_matrix形式
# 这样可以单独测试Process_data函数
a = CoraData()
a.process_data()
Out[3]:
Using Cached file: cora\processed_cora.pkl
Process data ...
test_index (1000,)
train_index (140,)
val_index (500,)
y_train_mask (140,)
num_nodesaaaaaaaaaaaa 2708
Node's feature shape: (2708, 1433)
Node's label shape: (2708,)
Adjacency's shape: (2708, 2708)
Number of training nodes: 140
Number of validation nodes: 500
Number of test nodes: 1000
Data(x=array([[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
...,
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.],
[0., 0., 0., ..., 0., 0., 0.]], dtype=float32), y=array([3, 4, 4, ..., 3, 3, 3], dtype=int64), adjacency=<2708x2708 sparse matrix of type '<class 'numpy.float32'>'
with 10556 stored elements in COOrdinate format>, train_mask=array([ True, True, True, ..., False, False, False]), val_mask=array([False, False, False, ..., False, False, False]), test_mask=array([False, False, False, ..., True, True, True]))
图卷积层定义
In [13]:
class GraphConvolution(nn.Module):
def __init__(self, input_dim, output_dim, use_bias=True):
"""图卷积:L*X*\theta
完整GCN函数
f = sigma(D^-1/2 A D^-1/2 * H * W)
卷积是D^-1/2 A D^-1/2 * H * W
adjacency = D^-1/2 A D^-1/2 已经经过归一化,标准化的拉普拉斯矩阵
这样就把傅里叶变化和拉普拉斯矩阵结合起来了.
Args:
----------
input_dim: int
节点输入特征的维度
output_dim: int
输出特征维度
use_bias : bool, optional
是否使用偏置
"""
super(GraphConvolution, self).__init__()
self.input_dim = input_dim
self.output_dim = output_dim
self.use_bias = use_bias
# 定义GCN层的 W 权重形状
self.weight = nn.Parameter(torch.Tensor(input_dim, output_dim))
#定义GCN层的 b 权重矩阵
if self.use_bias:
self.bias = nn.Parameter(torch.Tensor(output_dim))
else:
self.register_parameter('bias', None)
self.reset_parameters()
# 这里才是声明初始化 nn.Module 类里面的W,b参数
def reset_parameters(self):
init.kaiming_uniform_(self.weight)
if self.use_bias:
init.zeros_(self.bias)
def forward(self, adjacency, input_feature):
"""邻接矩阵是稀疏矩阵,因此在计算时使用稀疏矩阵乘法
Args:
-------
adjacency: torch.sparse.FloatTensor
邻接矩阵
input_feature: torch.Tensor
输入特征
"""
support = torch.mm(input_feature, self.weight) # 矩阵相乘 m由matrix缩写
output = torch.sparse.mm(adjacency, support) # sparse 稀疏的
if self.use_bias:
output += self.bias # bias 偏置,偏见
return output
# 一般是为了打印类实例的信息而重写的内置函数
def __repr__(self):
return self.__class__.__name__ + ' (' \
+ str(self.input_dim) + ' -> ' \
+ str(self.output_dim) + ')'
模型定义
读者可以自己对GCN模型结构进行修改和实验
In [14]:
class GcnNet(nn.Module):
"""
定义一个包含两层GraphConvolution的模型
"""
def __init__(self, input_dim=1433):
super(GcnNet, self).__init__()
self.gcn1 = GraphConvolution(input_dim, 16) #(N,1433)->(N,16)
self.gcn2 = GraphConvolution(16, 7) #(N,16)->(N,7)
def forward(self, adjacency, feature):
h = F.relu(self.gcn1(adjacency, feature)) #(N,1433)->(N,16),经过relu函数
logits = self.gcn2(adjacency, h) #(N,16)->(N,7)
return logits
模型训练
In [16]:
# 超参数定义
LEARNING_RATE = 0.1 #学习率
WEIGHT_DACAY = 5e-4 #正则化系数 weight_dacay
EPOCHS = 200 #完整遍历训练集的次数
DEVICE = "cuda" if torch.cuda.is_available() else "cpu" #设备
# 如果当前显卡忙于其他工作,可以设置为 DEVICE = "cpu",使用cpu运行
In [17]:
# 加载数据,并转换为torch.Tensor
dataset = CoraData().data
node_feature = dataset.x / dataset.x.sum(1, keepdims=True) # 归一化数据,使得每一行和为1
tensor_x = tensor_from_numpy(node_feature, DEVICE) # (2708,1433)
tensor_y = tensor_from_numpy(dataset.y, DEVICE) #(2708,)
tensor_train_mask = tensor_from_numpy(dataset.train_mask, DEVICE) #前140个为True
tensor_val_mask = tensor_from_numpy(dataset.val_mask, DEVICE) # 140 - 639 500个
tensor_test_mask = tensor_from_numpy(dataset.test_mask, DEVICE) # 1708 - 2707 1000个
normalize_adjacency = CoraData.normalization(dataset.adjacency) # 规范化邻接矩阵 计算 L=D^-0.5 * (A+I) * D^-0.5
num_nodes, input_dim = node_feature.shape # 2708,1433
# 原始创建coo_matrix((data, (row, col)), shape=(4, 4)) indices为index复数 https://blog.csdn.net/chao2016/article/details/80344828?ops_request_misc=%257B%2522request%255Fid%2522%253A%2522160509865819724838529777%2522%252C%2522scm%2522%253A%252220140713.130102334.pc%255Fall.%2522%257D&request_id=160509865819724838529777&biz_id=0&utm_medium=distribute.pc_search_result.none-task-blog-2~all~first_rank_v2~rank_v28-2-80344828.pc_first_rank_v2_rank_v28&utm_term=%E7%A8%80%E7%96%8F%E7%9F%A9%E9%98%B5%E7%9A%84coo_matrix&spm=1018.2118.3001.4449
indices = torch.from_numpy(np.asarray([normalize_adjacency.row,
normalize_adjacency.col]).astype('int64')).long()
values = torch.from_numpy(normalize_adjacency.data.astype(np.float32))
tensor_adjacency = torch.sparse.FloatTensor(indices, values,
(num_nodes, num_nodes)).to(DEVICE)
#根据三元组 构造 稀疏矩阵向量,构造出来的张量是 (2708,2708)
In [18]:
# 模型定义:Model, Loss, Optimizer
model = GcnNet(input_dim).to(DEVICE)
criterion = nn.CrossEntropyLoss().to(DEVICE) # criterion评判标准
optimizer = optim.Adam(model.parameters(), # optimizer 优化程序 ,使用Adam 优化方法,权重衰减https://blog.csdn.net/program_developer/article/details/80867468
lr=LEARNING_RATE,
weight_decay=WEIGHT_DACAY)
In [8]:
# 训练主体函数
def train():
loss_history = []
val_acc_history = []
model.train()
train_y = tensor_y[tensor_train_mask] # shape=(140,)不是(2708,)了
# 共进行200次训练
for epoch in range(EPOCHS):
logits = model(tensor_adjacency, tensor_x) # 前向传播,认为因为声明了 model.train(),不用forward了
train_mask_logits = logits[tensor_train_mask] # 只选择训练节点进行监督 (140,)
loss = criterion(train_mask_logits, train_y) # 计算损失值
optimizer.zero_grad() # 梯度归零
loss.backward() # 反向传播计算参数的梯度
optimizer.step() # 使用优化方法进行梯度更新
train_acc, _, _ = test(tensor_train_mask) # 计算当前模型训练集上的准确率 调用test函数
val_acc, _, _ = test(tensor_val_mask) # 计算当前模型在验证集上的准确率
# 记录训练过程中损失值和准确率的变化,用于画图
loss_history.append(loss.item())
val_acc_history.append(val_acc.item())
print("Epoch {:03d}: Loss {:.4f}, TrainAcc {:.4}, ValAcc {:.4f}".format(
epoch, loss.item(), train_acc.item(), val_acc.item()))
return loss_history, val_acc_history
In [9]:
# 测试函数
def test(mask):
model.eval() # 表示将模型转变为evaluation(测试)模式,这样就可以排除BN和Dropout对测试的干扰
with torch.no_grad(): # 显著减少显存占用
logits = model(tensor_adjacency, tensor_x) #(N,16)->(N,7) N节点数
test_mask_logits = logits[mask] # 矩阵形状和mask一样
predict_y = test_mask_logits.max(1)[1] # 返回每一行的最大值中索引(返回最大元素在各行的列索引)
accuarcy = torch.eq(predict_y, tensor_y[mask]).float().mean()
return accuarcy, test_mask_logits.cpu().numpy(), tensor_y[mask].cpu().numpy()
In [13]:
def plot_loss_with_acc(loss_history, val_acc_history):
fig = plt.figure()
# 坐标系ax1画曲线1
ax1 = fig.add_subplot(111) # 指的是将plot界面分成1行1列,此子图占据从左到右从上到下的1位置
ax1.plot(range(len(loss_history)), loss_history,
c=np.array([255, 71, 90]) / 255.) # c为颜色
plt.ylabel('Loss')
# 坐标系ax2画曲线2
ax2 = fig.add_subplot(111, sharex=ax1, frameon=False) # 其本质就是添加坐标系,设置共享ax1的x轴,ax2背景透明
ax2.plot(range(len(val_acc_history)), val_acc_history,
c=np.array([79, 179, 255]) / 255.)
ax2.yaxis.tick_right() # 开启右边的y坐标
ax2.yaxis.set_label_position("right")
plt.ylabel('ValAcc')
plt.xlabel('Epoch')
plt.title('Training Loss & Validation Accuracy')
plt.show()
In [ ]:
loss, val_acc = train()
test_acc, test_logits, test_label = test(tensor_test_mask)
print("Test accuarcy: ", test_acc.item())
In [14]:
plot_loss_with_acc(loss, val_acc)
。
。
。