二.GNN算法之Graph Network
2.1 GCN 图卷积神经网络
完整源码连接https://github.com/tkipf/pygcn
感谢作者!
from __future__ import division
from __future__ import print_function
import time
import argparse
import numpy as np
import torch
import torch.nn.functional as F
import torch.optim as optim
from pygcn.utils import load_data, accuracy
from pygcn.models import GCN
# 1.添加Training settings
parser = argparse.ArgumentParser()
parser.add_argument('--no-cuda', action='store_true', default=False,
help='Disables CUDA training.')
parser.add_argument('--fastmode', action='store_true', default=False,
help='Validate during training pass.')
parser.add_argument('--seed', type=int, default=42, help='Random seed.')
parser.add_argument('--epochs', type=int, default=200,
help='Number of epochs to train.')
parser.add_argument('--lr', type=float, default=0.01,
help='Initial learning rate.')
parser.add_argument('--weight_decay', type=float, default=5e-4,
help='Weight decay (L2 loss on parameters).')
parser.add_argument('--hidden', type=int, default=16,
help='Number of hidden units.')
parser.add_argument('--dropout', type=float, default=0.5,
help='Dropout rate (1 - keep probability).')
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# MAC: option + command + <-
# Load data2.导入数据集
adj, features, labels, idx_train, idx_val, idx_test = load_data()
# Model and optimizer,构造GCN,初始化参数。两层GCN
model = GCN(nfeat=features.shape[1],
nhid=args.hidden,
nclass=labels.max().item() + 1,
dropout=args.dropout)
optimizer = optim.Adam(model.parameters(),
lr=args.lr, weight_decay=args.weight_decay)#优化器
#由GPU的情况下:
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()
#定义训练模型
def train(epoch):
t = time.time()
model.train()
optimizer.zero_grad() # GraphConvolution forward
output = model(features, adj) # 运行模型,输入参数 (features, adj)
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.item()),
'acc_train: {:.4f}'.format(acc_train.item()),
'loss_val: {:.4f}'.format(loss_val.item()),
'acc_val: {:.4f}'.format(acc_val.item()),
'time: {:.4f}s'.format(time.time() - t))
#测试模型
def 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.item()),
"accuracy= {:.4f}".format(acc_test.item()))
# Train model
t_total = time.time()
for epoch in range(args.epochs):
train(epoch)
print("Optimization Finished!")
print("Total time elapsed: {:.4f}s".format(time.time() - t_total))
# Testing
test()
创建一个两层的GCN模型
class GCN(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout):
super(GCN, self).__init__()
self.gc1 = GraphConvolution(nfeat, nhid) # 构建第一层 GCN(初始特征,隐藏层特征)
self.gc2 = GraphConvolution(nhid, nclass) # 构建第二层 GCN
self.dropout = dropout
def forward(self, x, adj):
x = F.relu(self.gc1(x, adj))
x = F.dropout(x, self.dropout, training=self.training)
x = self.gc2(x, adj)
return F.log_softmax(x, dim=1)
import math
import torch
from torch.nn.parameter import Parameter
from torch.nn.modules.module import Module
class GraphConvolution(Module):
"""
Simple GCN layer, similar to https://arxiv.org/abs/1609.02907
"""
def __init__(self, in_features, out_features, bias=True):
super(GraphConvolution, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.weight = Parameter(torch.FloatTensor(in_features, out_features)) # 构造weight函数,输入输出为input_features, out_features
if bias:
self.bias = Parameter(torch.FloatTensor(out_features))
else:
self.register_parameter('bias', None)
self.reset_parameters()
def reset_parameters(self):
stdv = 1. / math.sqrt(self.weight.size(1))
self.weight.data.uniform_(-stdv, stdv) # 随机化参数
if self.bias is not None:
self.bias.data.uniform_(-stdv, stdv)
def forward(self, input, adj):
support = torch.mm(input, self.weight) # GraphConvolution forward。input*weight
output = torch.spmm(adj, support) # 稀疏矩阵的相乘,和mm一样的效果
if self.bias is not None:
return output + self.bias
else:
return output
def __repr__(self):
return self.__class__.__name__ + ' (' \
+ str(self.in_features) + ' -> ' \
+ str(self.out_features) + ')'
2.2 GAT图注意力模型
完整源码连接https://github.com/tkipf/pygat
感谢作者!
from __future__ import division
from __future__ import print_function
import os
import glob
import time
import random
import argparse
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from utils import load_data, accuracy
from models import GAT, SpGAT
# Training settings
parser = argparse.ArgumentParser()
parser.add_argument('--no-cuda', action='store_true', default=False, help='Disables CUDA training.')
parser.add_argument('--fastmode', action='store_true', default=False, help='Validate during training pass.')
parser.add_argument('--sparse', action='store_true', default=False, help='GAT with sparse version or not.')
parser.add_argument('--seed', type=int, default=72, help='Random seed.')
parser.add_argument('--epochs', type=int, default=10000, help='Number of epochs to train.')
parser.add_argument('--lr', type=float, default=0.005, help='Initial learning rate.')
parser.add_argument('--weight_decay', type=float, default=5e-4, help='Weight decay (L2 loss on parameters).')
parser.add_argument('--hidden', type=int, default=8, help='Number of hidden units.')
parser.add_argument('--nb_heads', type=int, default=8, help='Number of head attentions.')
parser.add_argument('--dropout', type=float, default=0.6, help='Dropout rate (1 - keep probability).')
parser.add_argument('--alpha', type=float, default=0.2, help='Alpha for the leaky_relu.')
parser.add_argument('--patience', type=int, default=100, help='Patience')
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
if args.cuda:
torch.cuda.manual_seed(args.seed)
# Load data
adj, features, labels, idx_train, idx_val, idx_test = load_data()
# Model and optimizer
args.sparse = True
#构建GAT模型
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)
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)
def train(epoch):
t = time.time()
model.train()
optimizer.zero_grad()
output = model(features, adj) # GAT模块
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[0]),
"accuracy= {:.4f}".format(acc_test.data[0]))
# 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()
import torch
import torch.nn as nn
import torch.nn.functional as F
from layers import GraphAttentionLayer, SpGraphAttentionLayer
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)]#原始特征,隐含特征,nheads多头向量=8
for i, attention in enumerate(self.attentions):
self.add_module('attention_{}'.format(i), attention)
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) # 将每层attention拼接
x = F.dropout(x, self.dropout, training=self.training)
x = F.elu(self.out_att(x, adj)) # 第二层的attention layer
return F.log_softmax(x, dim=1)
class SpGAT(nn.Module):
def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
"""Sparse version of GAT."""
super(SpGAT, self).__init__()
self.dropout = dropout
self.attentions = [SpGraphAttentionLayer(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)
self.out_att = SpGraphAttentionLayer(nhid * nheads,
nclass,
dropout=dropout,
alpha=alpha,
concat=False)
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)
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
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
#w,a向量(基于公式)
self.W = nn.Parameter(torch.empty(size=(in_features, out_features)))
nn.init.xavier_uniform_(self.W.data, gain=1.414)
self.a = nn.Parameter(torch.empty(size=(2*out_features, 1))) # concat(V,NeigV)
nn.init.xavier_uniform_(self.a.data, gain=1.414)
self.leakyrelu = nn.LeakyReLU(self.alpha)
def forward(self, h, adj):#公式
Wh = torch.mm(h, self.W) # h.shape: (N, in_features), Wh.shape: (N, out_features)
a_input = self._prepare_attentional_mechanism_input(Wh) # 每一个节点和所有节点,特征。(Vall, Vall, feature)
e = self.leakyrelu(torch.matmul(a_input, self.a).squeeze(2))
# 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)
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
attention = F.dropout(attention, self.dropout, training=self.training)
h_prime = torch.matmul(attention, Wh) # 聚合邻居函数,得到节点的最终表示
if self.concat:
return F.elu(h_prime) # elu-激活函数
else:
return h_prime
def _prepare_attentional_mechanism_input(self, Wh):
N = Wh.size()[0] # number of nodes
# Below, two matrices are created that contain embeddings in their rows in different orders.
# (e stands for embedding)
# These are the rows of the first matrix (Wh_repeated_in_chunks):
# e1, e1, ..., e1, e2, e2, ..., e2, ..., eN, eN, ..., eN
# '-------------' -> N times '-------------' -> N times '-------------' -> N times
#
# These are the rows of the second matrix (Wh_repeated_alternating):
# e1, e2, ..., eN, e1, e2, ..., eN, ..., e1, e2, ..., eN
# '----------------------------------------------------' -> N times
#
Wh_repeated_in_chunks = Wh.repeat_interleave(N, dim=0) # 复制
Wh_repeated_alternating = Wh.repeat(N, 1)
# Wh_repeated_in_chunks.shape == Wh_repeated_alternating.shape == (N * N, out_features)
# The all_combination_matrix, created below, will look like this (|| denotes concatenation):
# e1 || e1
# e1 || e2
# e1 || e3
# ...
# e1 || eN
# e2 || e1
# e2 || e2
# e2 || e3
# ...
# e2 || eN
# ...
# eN || e1
# eN || e2
# eN || e3
# ...
# eN || eN
all_combinations_matrix = torch.cat([Wh_repeated_in_chunks, Wh_repeated_alternating], dim=1)
# all_combinations_matrix.shape == (N * N, 2 * out_features)
return all_combinations_matrix.view(N, N, 2 * self.out_features)
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features) + ' -> ' + str(self.out_features) + ')'
class SpecialSpmmFunction(torch.autograd.Function):
"""Special function for only sparse region backpropataion layer."""
@staticmethod
def forward(ctx, indices, values, shape, b):
assert indices.requires_grad == False
a = torch.sparse_coo_tensor(indices, values, shape)
ctx.save_for_backward(a, b)
ctx.N = shape[0]
return torch.matmul(a, b)
@staticmethod
def backward(ctx, grad_output):
a, b = ctx.saved_tensors
grad_values = grad_b = None
if ctx.needs_input_grad[1]:
grad_a_dense = grad_output.matmul(b.t())
edge_idx = a._indices()[0, :] * ctx.N + a._indices()[1, :]
grad_values = grad_a_dense.view(-1)[edge_idx]
if ctx.needs_input_grad[3]:
grad_b = a.t().matmul(grad_output)
return None, grad_values, None, grad_b
class SpecialSpmm(nn.Module):
def forward(self, indices, values, shape, b):
return SpecialSpmmFunction.apply(indices, values, shape, b)
class SpGraphAttentionLayer(nn.Module):
"""
Sparse version GAT layer, similar to https://arxiv.org/abs/1710.10903
"""
def __init__(self, in_features, out_features, dropout, alpha, concat=True):
super(SpGraphAttentionLayer, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.alpha = alpha
self.concat = concat
self.W = nn.Parameter(torch.zeros(size=(in_features, out_features)))
nn.init.xavier_normal_(self.W.data, gain=1.414)
self.a = nn.Parameter(torch.zeros(size=(1, 2*out_features)))
nn.init.xavier_normal_(self.a.data, gain=1.414)
self.dropout = nn.Dropout(dropout)
self.leakyrelu = nn.LeakyReLU(self.alpha)
self.special_spmm = SpecialSpmm()
def forward(self, input, adj):
dv = 'cuda' if input.is_cuda else 'cpu'
N = input.size()[0]
edge = adj.nonzero().t()
h = torch.mm(input, self.W)
# h: N x out
assert not torch.isnan(h).any()
# Self-attention on the nodes - Shared attention mechanism
edge_h = torch.cat((h[edge[0, :], :], h[edge[1, :], :]), dim=1).t()
# edge: 2*D x E
edge_e = torch.exp(-self.leakyrelu(self.a.mm(edge_h).squeeze()))
assert not torch.isnan(edge_e).any()
# edge_e: E
e_rowsum = self.special_spmm(edge, edge_e, torch.Size([N, N]), torch.ones(size=(N,1), device=dv))
# e_rowsum: N x 1
edge_e = self.dropout(edge_e)
# edge_e: E
h_prime = self.special_spmm(edge, edge_e, torch.Size([N, N]), h)
assert not torch.isnan(h_prime).any()
# h_prime: N x out
h_prime = h_prime.div(e_rowsum)
# h_prime: N x out
assert not torch.isnan(h_prime).any()
if self.concat:
# if this layer is not last layer,
return F.elu(h_prime)
else:
# if this layer is last layer,
return h_prime
def __repr__(self):
return self.__class__.__name__ + ' (' + str(self.in_features) + ' -> ' + str(self.out_features) + ')'
2.3 GraphSAGE
基本思想:
去学习一个节点的信息是怎么通过其邻居节点的特征聚合而来的。 学习到了这样的“聚合函数”,而我们本身就已知各个节点的特征和邻居关系,我们就可以很方便地得到一个新节点的表示了。
main
完整源码连接https://github.com/tkipf/
感谢作者!
import sys
import os
import torch
import argparse
import pyhocon
import random
from src.dataCenter import *
from src.utils import *
from src.models import *
parser = argparse.ArgumentParser(description='pytorch version of GraphSAGE')
parser.add_argument('--dataSet', type=str, default='cora')
parser.add_argument('--agg_func', type=str, default='MEAN')
parser.add_argument('--epochs', type=int, default=50)
parser.add_argument('--b_sz', type=int, default=20)
parser.add_argument('--seed', type=int, default=824)
parser.add_argument('--cuda', action='store_true',
help='use CUDA')
parser.add_argument('--gcn', action='store_true')
parser.add_argument('--learn_method', type=str, default='sup')
parser.add_argument('--unsup_loss', type=str, default='normal')
parser.add_argument('--max_vali_f1', type=float, default=0)
parser.add_argument('--name', type=str, default='debug')
parser.add_argument('--config', type=str, default='./experiments.conf') #/src
args = parser.parse_args()
if torch.cuda.is_available():
if not args.cuda:
print("WARNING: You have a CUDA device, so you should probably run with --cuda")
else:
device_id = torch.cuda.current_device()
print('using device', device_id, torch.cuda.get_device_name(device_id))
device = torch.device("cuda" if args.cuda else "cpu")
print('DEVICE:', device)
if __name__ == '__main__':
random.seed(args.seed)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed_all(args.seed)
# load config file
config = pyhocon.ConfigFactory.parse_file(args.config)
# load data读取数据
ds = args.dataSet
dataCenter = DataCenter(config)
dataCenter.load_dataSet(ds) # 读取数据
features = torch.FloatTensor(getattr(dataCenter, ds+'_feats')).to(device)
graphSage = GraphSage(config['setting.num_layers'], features.size(1), config['setting.hidden_emb_size'], features, getattr(dataCenter, ds+'_adj_lists'), device, gcn=args.gcn, agg_func=args.agg_func)
graphSage.to(device)
num_labels = len(set(getattr(dataCenter, ds+'_labels'))) # label的数量
classification = Classification(config['setting.hidden_emb_size'], num_labels)#分类器
classification.to(device)#分类器转换到cpu或gpu
#无监督的loss
unsupervised_loss = UnsupervisedLoss(getattr(dataCenter, ds+'_adj_lists'), getattr(dataCenter, ds+'_train'), device)
if args.learn_method == 'sup':#如果学习的是有监督的任务
print('GraphSage with Supervised Learning')
elif args.learn_method == 'plus_unsup':
print('GraphSage with Supervised Learning plus Net Unsupervised Learning')
else:
print('GraphSage with Net Unsupervised Learning')
#开始训练
for epoch in range(args.epochs):
print('----------------------EPOCH %d-----------------------' % epoch)
# 返回的是模型
graphSage, classification = apply_model(dataCenter, ds, graphSage, classification, unsupervised_loss, args.b_sz, args.unsup_loss, device, args.learn_method)
if (epoch+1) % 2 == 0 and args.learn_method == 'unsup':
classification, args.max_vali_f1 = train_classification(dataCenter, graphSage, classification, ds, device, args.max_vali_f1, args.name)
if args.learn_method != 'unsup':
args.max_vali_f1 = evaluate(dataCenter, ds, graphSage, classification, device, args.max_vali_f1, args.name, epoch)
module
class GraphSage(nn.Module):
"""docstring for GraphSage"""
def __init__(self, num_layers, input_size, out_size, raw_features, adj_lists, device, gcn=False, agg_func='MEAN'):
super(GraphSage, self).__init__()
self.input_size = input_size
self.out_size = out_size
self.num_layers = num_layers
self.gcn = gcn#是否使用gcn
self.device = device
self.agg_func = agg_func#聚合函数
self.raw_features = raw_features # 点的特征
self.adj_lists = adj_lists # 边的连接
for index in range(1, num_layers+1):
layer_size = out_size if index != 1 else input_size
setattr(self, 'sage_layer'+str(index), SageLayer(layer_size, out_size, gcn=self.gcn))
def forward(self, nodes_batch):
"""
Generates embeddings for a batch of nodes.
nodes_batch -- batch of nodes to learn the embeddings. 《minbatch 过程,涉及到的所有节点》
"""
lower_layer_nodes = list(nodes_batch)
nodes_batch_layers = [(lower_layer_nodes,)] # 第一次放入的节点,batch节点
# self.dc.logger.info('get_unique_neighs.')
for i in range(self.num_layers): # 每层的Sage
lower_samp_neighs, lower_layer_nodes_dict, lower_layer_nodes= self._get_unique_neighs_list(lower_layer_nodes) # 获得neighbors。 聚合自己和邻居节点,点的dict,涉及到的所有节点
nodes_batch_layers.insert(0, (lower_layer_nodes, lower_samp_neighs, lower_layer_nodes_dict)) # 聚合自己和邻居节点,点的dict,涉及到的所有节点
# insert,0 从最外层开始聚合
assert len(nodes_batch_layers) == self.num_layers + 1
pre_hidden_embs = self.raw_features
for index in range(1, self.num_layers+1):
nb = nodes_batch_layers[index][0] # 聚合自己和周围的节点
pre_neighs = nodes_batch_layers[index-1] # 这层节点的上层邻居的所有信息。聚合自己和邻居节点,点的dict,涉及到的所有节点
# self.dc.logger.info('aggregate_feats.') aggrefate_feats=>输出GraphSAGE聚合后的信息
aggregate_feats = self.aggregate(nb, pre_hidden_embs, pre_neighs) # 聚合函数。nb-这一层的节点, pre_hidden_embs-feature,pre_neighs-上一层节点
sage_layer = getattr(self, 'sage_layer'+str(index))
if index > 1:
nb = self._nodes_map(nb, pre_hidden_embs, pre_neighs) # 第一层的batch节点,没有进行转换
# self.dc.logger.info('sage_layer.')
cur_hidden_embs = sage_layer(self_feats=pre_hidden_embs[nb],
aggregate_feats=aggregate_feats) # 进入SageLayer。weight*concat(node,neighbors)
pre_hidden_embs = cur_hidden_embs
return pre_hidden_embs