由于和GCN代码比较相似,所以部分内容从GCN那篇博客中截取。
1 - cora数据集
2 - 源码含义记录
首先我们来整体看一下代码的组成
截图中的这一大坨为命令行传递参数,含义参考命令行传递参数 argparse.ArgumentParser解析简单点说,就是想在不改动代码的情况下,使用命令行去改参数。
2.1 加载数据集
在代码中,加载数据集通过这个函数实现
# Load data
adj, features, labels, idx_train, idx_val, idx_test = load_data()
def load_data(path="./data/cora/", dataset="cora"):
"""Load citation network dataset (cora only for now)"""
print('Loading {} dataset...'.format(dataset))
idx_features_labels = np.genfromtxt("{}{}.content".format(path, dataset), dtype=np.dtype(str))
features = sp.csr_matrix(idx_features_labels[:, 1:-1], dtype=np.float32)
labels = encode_onehot(idx_features_labels[:, -1])
# build graph
idx = np.array(idx_features_labels[:, 0], dtype=np.int32)
idx_map = {j: i for i, j in enumerate(idx)}
edges_unordered = np.genfromtxt("{}{}.cites".format(path, dataset), dtype=np.int32)
edges = np.array(list(map(idx_map.get, edges_unordered.flatten())), dtype=np.int32).reshape(edges_unordered.shape)
adj = sp.coo_matrix((np.ones(edges.shape[0]), (edges[:, 0], edges[:, 1])), shape=(labels.shape[0], labels.shape[0]), dtype=np.float32)
# build symmetric adjacency matrix
adj = adj + adj.T.multiply(adj.T > adj) - adj.multiply(adj.T > adj)
features = normalize_features(features)
adj = normalize_adj(adj + sp.eye(adj.shape[0]))
idx_train = range(140)
idx_val = range(200, 500)
idx_test = range(500, 1500)
adj = torch.FloatTensor(np.array(adj.todense()))
features = torch.FloatTensor(np.array(features.todense()))
labels = torch.LongTensor(np.where(labels)[1])
idx_train = torch.LongTensor(idx_train)
idx_val = torch.LongTensor(idx_val)
idx_test = torch.LongTensor(idx_test)
return adj, features, labels, idx_train, idx_val, idx_test
np.genfromtxt
函数为numpy加载数据集,当然还有其它几种加载数据集的方式,例如pandas等等,详情见Python加载数据的5种不同方式(收藏)。
2.1.1 加载节点数据
首先是加载节点的数据,即.content。
通过断点+命令行调试,可以看到输出的idx_features_labels
的结果
因此我们想把中间节点的特征
给取出来,即这一句
features = sp.csr_matrix(idx_features_labels[:, 1:-1], dtype=np.float32)
中间[1:-1]
是从第二列到倒数第二列(因为python为左闭右开)
然后再取label
,使用
labels = encode_onehot(idx_features_labels[:, -1])
只取最后一列,并且采用one-hot编码。
分别看看上述两行代码的输出
解释一下特征的含义
最后开始取节点的索引(index
),即第一列,并且构建节点的索引字典。
idx = np.array(idx_features_labels[:, 0], dtype=np.int32) # 取节点
idx_map = {j: i for i, j in enumerate(idx)} # 构建节点的索引字典
看看输出,字典就构建好啦
2.1.2 加载边的数据
节点的数据导入完了,我们再来导入边的数据
edges_unordered = np.genfromtxt("{}{}.cites".format(path, dataset), dtype=np.int32) # 导入edge的数据
将之前的转换成字典编号后的边
2.1.3 构造邻接矩阵
adj = sp.coo_matrix((np.ones(edges.shape[0]), (edges[:, 0], edges[:, 1])),
shape=(labels.shape[0], labels.shape[0]),
dtype=np.float32) # 构造边的邻接矩阵(稀疏矩阵)
看看邻接矩阵的大小和样子,因为有2708个节点,所以大小为(2708×2708)
由于Cora数据集是一个有向图,而GCN本身使用的为无向图,其邻接矩阵是对称的,因此我们需要构造一个对称矩阵,如下:
# build symmetric adjacency matrix 计算邻接矩阵的对称矩阵,因为cora是有向图,而GCN是无向图
adj = adj + adj.T.multiply(adj.T > adj) - adj.multiply(adj.T > adj)
其中adj.T > adj
的含义为,如果转置后为1,而原矩阵为0,则该位置为1
Python将非对称邻接矩阵转变为对称邻接矩阵(有向图转无向图)
2.1.4 归一化
features = normalize(features)
adj = normalize(adj + sp.eye(adj.shape[0]))
我们可以看一下normalize
函数
def normalize(mx):
"""Row-normalize sparse matrix"""
rowsum = np.array(mx.sum(1)) # 矩阵行求和
r_inv = np.power(rowsum, -1).flatten() # 求和的-1次方,即倒数
r_inv[np.isinf(r_inv)] = 0. # 求倒数的时候有可能是无穷,因此如果是无穷的话,就转换成0
r_mat_inv = sp.diags(r_inv) # 构造对角矩阵
mx = r_mat_inv.dot(mx) # 构造D-1 * A,特征归一化,非对称方式,简化方式
return mx
分别看看rowsum
和r_inv
的输出
然后我们在把r_inv中的无穷值变为0后,构造对角矩阵,可以看到结果
特征矩阵归一化(非必须)
做了上述这些后,我们对特征矩阵进行归一化
可以发现特征每一行加起来都是1,已经被归一化好了。
邻接矩阵归一化(必须)
再对邻接矩阵+单位阵进行归一化(必须操作)
首先看起来邻接矩阵+单位阵的值:
然后再进行归一化,这里用的是简便方法,和上面一样啦~
2.1.5 加载数据集/数据格式转换
# 训练、验证、测试的样本
idx_train = range(140)
idx_val = range(200, 500)
idx_test = range(500, 1500)
# 将numpy的数据转换成torch格式
features = torch.FloatTensor(np.array(features.todense()))
labels = torch.LongTensor(np.where(labels)[1])
adj = sparse_mx_to_torch_sparse_tensor(adj)
idx_train = torch.LongTensor(idx_train)
idx_val = torch.LongTensor(idx_val)
最后返回的格式
return adj, features, labels, idx_train, idx_val, idx_test
2.2 构造模型
我们构造一个GAT模型和优化器
# Model and optimizer
if args.sparse:
model = SpGAT(nfeat=features.shape[1],
nhid=args.hidden,
nclass=int(labels.max()) + 1,
dropout=args.dropout,
nheads=args.nb_heads,
alpha=args.alpha)
else:
model = GAT(nfeat=features.shape[1],
nhid=args.hidden,
nclass=int(labels.max()) + 1,
dropout=args.dropout,
nheads=args.nb_heads,
alpha=args.alpha)
optimizer = optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
看一下GAT长什么样
class GAT(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
"""Dense version of GAT."""
super(GAT, self).__init__()
self.dropout = dropout
self.attentions = [GraphAttentionLayer(nfeat, nhid, dropout=dropout, alpha=alpha, concat=True) for _ in range(nheads)]
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention) # 这里就有8个attention layer,第一层的attention layer
self.out_att = GraphAttentionLayer(nhid * nheads, nclass, dropout=dropout, alpha=alpha, concat=False) # 第二层(最后一层)的attention layer
def forward(self, x, adj):
x = F.dropout(x, self.dropout, training=self.training)
x = torch.cat([att(x, adj) for att in self.attentions], dim=1)
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, adj))
return F.log_softmax(x, dim=1)
首先设置dropout,然后比较重要的就是接下来的
[GraphAttentionLayer(nfeat, nhid, dropout=dropout, alpha=alpha, concat=True) for _ in range(nheads)]
通过断点调试,可以看到上述函数中几个参数的值以及含义
- nfeat:输入特征=1433
- nhid:隐藏特征=8
- nheads:多头向量=8,代表有8个attentionlayer
多头操作如下:
了解了GraphAttentionLayer
的输入参数,再来看看具体定义:
class GraphAttentionLayer(nn.Module):
"""
Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, concat=True):
super(GraphAttentionLayer, self).__init__()
self.dropout = dropout
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
self.W = nn.Parameter(torch.empty(size=(in_features, out_features))) # in_features=1433, out_features=8
nn.init.xavier_uniform_(self.W.data, gain=1.414) # 初始化W向量
self.a = nn.Parameter(torch.empty(size=(2*out_features, 1))) # 建立一个需要训练的α向量,因为是将两个向量Whi和Whj拼接到一起,所以维度是2×out_features self.a.shape = torch.Size([16, 1])
nn.init.xavier_uniform_(self.a.data, gain=1.414) # 初始化α向量,赋予随机的值
self.leakyrelu = nn.LeakyReLU(self.alpha)
def forward(self, h, adj):
Wh = torch.mm(h, self.W) # h.shape: (2708, 1433) self.W.shape (1433.8) W*h = Wh
e = self._prepare_attentional_mechanism_input(Wh) # 每一个节点和所有节点,特征拼接?
# 之前计算的是一个节点和所有节点的attention,其实需要的是连接的节点的attention系数
zero_vec = -9e15*torch.ones_like(e) # 建立了一个极小值矩阵
attention = torch.where(adj > 0, e, zero_vec) # 将邻接矩阵中小于0的变成负无穷
attention = F.softmax(attention, dim=1) # 按行求softmax,sum(axis=1) === 1 归一化操作
attention = F.dropout(attention, self.dropout, training=self.training)
h_prime = torch.matmul(attention, Wh) # 聚合邻居函数
if self.concat:
return F.elu(h_prime)
else:
return h_prime
def _prepare_attentional_mechanism_input(self, Wh):
# Wh.shape (N, out_feature)
# self.a.shape (2 * out_feature, 1)
# Wh1&2.shape (N, 1)
# e.shape (N, N)
Wh1 = torch.matmul(Wh, self.a[:self.out_features, :])
Wh2 = torch.matmul(Wh, self.a[self.out_features:, :])
# broadcast add
e = Wh1 + Wh2.T
return self.leakyrelu(e)
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features) + ' -> ' + str(self.out_features) + ')'
2.3 使用GPU
if args.cuda:
model.cuda()
features = features.cuda()
adj = adj.cuda()
labels = labels.cuda()
idx_train = idx_train.cuda()
idx_val = idx_val.cuda()
idx_test = idx_test.cuda()
features, adj, labels = Variable(features), Variable(adj), Variable(labels)
2.4 定义训练和测试函数
def train(epoch):
t = time.time()
model.train()
optimizer.zero_grad()
output = model(features, adj) # features [2708, 1433]
loss_train = F.nll_loss(output[idx_train], labels[idx_train])
acc_train = accuracy(output[idx_train], labels[idx_train])
loss_train.backward()
optimizer.step()
if not args.fastmode:
# Evaluate validation set performance separately,
# deactivates dropout during validation run.
model.eval()
output = model(features, adj)
loss_val = F.nll_loss(output[idx_val], labels[idx_val])
acc_val = accuracy(output[idx_val], labels[idx_val])
print('Epoch: {:04d}'.format(epoch+1),
'loss_train: {:.4f}'.format(loss_train.data.item()),
'acc_train: {:.4f}'.format(acc_train.data.item()),
'loss_val: {:.4f}'.format(loss_val.data.item()),
'acc_val: {:.4f}'.format(acc_val.data.item()),
'time: {:.4f}s'.format(time.time() - t))
return loss_val.data.item()
def compute_test():
model.eval()
output = model(features, adj)
loss_test = F.nll_loss(output[idx_test], labels[idx_test])
acc_test = accuracy(output[idx_test], labels[idx_test])
print("Test set results:",
"loss= {:.4f}".format(loss_test.data.item()),
"accuracy= {:.4f}".format(acc_test.data.item()))
2.5 模型训练与测试
# Train model
t_total = time.time()
loss_values = []
bad_counter = 0
best = args.epochs + 1
best_epoch = 0
for epoch in range(args.epochs):
loss_values.append(train(epoch))
torch.save(model.state_dict(), '{}.pkl'.format(epoch))
if loss_values[-1] < best:
best = loss_values[-1]
best_epoch = epoch
bad_counter = 0
else:
bad_counter += 1
if bad_counter == args.patience:
break
files = glob.glob('*.pkl')
for file in files:
epoch_nb = int(file.split('.')[0])
if epoch_nb < best_epoch:
os.remove(file)
files = glob.glob('*.pkl')
for file in files:
epoch_nb = int(file.split('.')[0])
if epoch_nb > best_epoch:
os.remove(file)
print("Optimization Finished!")
print("Total time elapsed: {:.4f}s".format(time.time() - t_total))
# Restore best model
print('Loading {}th epoch'.format(best_epoch))
model.load_state_dict(torch.load('{}.pkl'.format(best_epoch)))
# Testing
compute_test()