def pretreatment(self, data, test=False):
'''
将从dataloader加载出来的data转化为可以传入神经网络的数据
'''
import torch
from torch.autograd import Variable
from utils.loadData import loadData
image = torch.FloatTensor(self.opt.MODEL.BATCH_SIZE,
self.opt.IMAGE.IMG_CHANNEL,
self.opt.IMAGE.IMG_H, self.opt.IMAGE.IMG_H)
text = torch.LongTensor(self.opt.MODEL.BATCH_SIZE * 5)
text_rev = torch.LongTensor(self.opt.MODEL.BATCH_SIZE * 5)
length = torch.IntTensor(self.opt.MODEL.BATCH_SIZE)
if self.opt.BASE.CUDA:
# self.model = torch.nn.DataParallel(self.model, device_ids=range(self.opt.ngpu))
image = image.cuda()
text = text.cuda()
text_rev = text_rev.cuda()
self.criterion = self.criterion.cuda()
image = Variable(image)
text = Variable(text)
text_rev = Variable(text_rev)
length = Variable(length)
cpu_images, cpu_texts = data
loadData(image, cpu_images)
t, l = self.converter.encode(cpu_texts, scanned=True)
loadData(text, t)
loadData(length, l)
return image, length, text, text_rev, test
# 网络训练网络文件名
trainName = "bitcoinAlpha" # bitcoinAlpha bitcoinOTC epinions_truncated slashdot_truncated
# 网络经过tsvd分解的特征大小 默认为64
int_features = 64
# 模型
modelName = "GCN" # GCN or HGCN
# 网络卷积后的特征大小 默认为64
out_features = 64
# 卷积层大小,即1次邻居卷积 加num_layers-1层间接邻居卷积
num_layers = 2
# loss 参数
lambda_structure = 5
# 学习率
lr = 0.01
pos_edge_index, neg_edge_index = loadData(trainFiles, trainName)
train_pos_edge_index, test_pos_edge_index = split_edges(pos_edge_index)
train_neg_edge_index, test_neg_edge_index = split_edges(neg_edge_index)
# 自训练向量大小
posAtt = train_pos_edge_index.shape[1]
negAtt = train_neg_edge_index.shape[1]
model = None
if modelName == "HGCN":
model = SignHGCN(int_features,
out_features,
num_layers=num_layers,
lambda_structure=lambda_structure)
if modelName == "GCN":
model = SignGCN(int_features,
out_features,