GraphSAGE_RL:通过数据驱动的节点采样提高GraphSAGE

上传者: 42148975 | 上传时间: 2022-04-16 22:59:33 | 文件大小: 7.29MB | 文件类型: ZIP
通过数据驱动的节点采样提高GraphSAGE 作者: ( , ), ( ), ( ) 发表在2019 ICLR研讨会表示学习上图和流形。 概述 作为一种高效且可扩展的图神经网络,GraphSAGE通过归纳二次采样的局部邻域并以小批量梯度下降的方式进行学习,已启用了归纳能力来推断看不见的节点或图。 GraphSAGE中使用的邻域采样有效地提高了并行推断一批不同程度的目标节点时的计算和存储效率。 尽管有此优势,但默认的统一采样在训练和推理上仍存在较大差异,从而导致次优准确性。 我们提出了一种新的数据驱动的采样方法,以通过非线性回归来推断邻域的实际值重要性,并使用该值作为对邻域进行二次采样的标准。 使用基于值的强化学习来学习回归者。 从GraphSAGE的负分类损失输出中归纳地提取了顶点和邻域的每种组合的隐含重要性。 结果,在使用三个数据集的归纳节点分类基准中,我们的方法使用统一

文件下载

资源详情

[{"title":"( 40 个子文件 7.29MB ) GraphSAGE_RL:通过数据驱动的节点采样提高GraphSAGE","children":[{"title":"GraphSAGE_RL-master","children":[{"title":"Dockerfile.gpu <span style='color:#111;'> 108B </span>","children":null,"spread":false},{"title":"utils.py <span style='color:#111;'> 10.14KB </span>","children":null,"spread":false},{"title":"eval_scripts","children":[{"title":"ppi_eval.py <span style='color:#111;'> 3.68KB </span>","children":null,"spread":false},{"title":"citation_eval.py <span style='color:#111;'> 4.43KB </span>","children":null,"spread":false},{"title":"reddit_eval.py <span style='color:#111;'> 4.69KB </span>","children":null,"spread":false}],"spread":true},{"title":"data","children":[{"title":"pubmed","children":[{"title":"pubmed-feats.npy <span style='color:#111;'> 37.61MB </span>","children":null,"spread":false},{"title":"pubmed-id_map.json <span style='color:#111;'> 286.38KB </span>","children":null,"spread":false},{"title":"pubmed-class_map.json <span style='color:#111;'> 220.21KB </span>","children":null,"spread":false},{"title":"pubmed-G.json <span style='color:#111;'> 3.12MB </span>","children":null,"spread":false}],"spread":true}],"spread":true},{"title":"model","children":[{"title":".gitkeep <span style='color:#111;'> 2B </span>","children":null,"spread":false}],"spread":true},{"title":"create_Graph_forGraphSAGE.py <span style='color:#111;'> 1.59KB </span>","children":null,"spread":false},{"title":"output","children":[{"title":".gitkeep <span style='color:#111;'> 2B </span>","children":null,"spread":false}],"spread":true},{"title":"example_supervised.sh <span style='color:#111;'> 650B </span>","children":null,"spread":false},{"title":"requirements.txt <span style='color:#111;'> 397B </span>","children":null,"spread":false},{"title":".gitignore <span style='color:#111;'> 1.14KB </span>","children":null,"spread":false},{"title":"loss_node","children":[{"title":".gitkeep <span style='color:#111;'> 2B </span>","children":null,"spread":false}],"spread":true},{"title":"README.md <span style='color:#111;'> 6.81KB </span>","children":null,"spread":false},{"title":"example_data","children":[{"title":"toy-ppi-id_map.json <span style='color:#111;'> 208.85KB </span>","children":null,"spread":false},{"title":"toy-ppi-class_map.json <span style='color:#111;'> 5.25MB </span>","children":null,"spread":false},{"title":"toy-ppi-G.json <span style='color:#111;'> 27.20MB </span>","children":null,"spread":false},{"title":"toy-ppi-feats.npy <span style='color:#111;'> 5.63MB </span>","children":null,"spread":false}],"spread":true},{"title":"graphsage","children":[{"title":"tags <span style='color:#111;'> 25.99KB </span>","children":null,"spread":false},{"title":"neigh_samplers.py <span style='color:#111;'> 8.30KB </span>","children":null,"spread":false},{"title":"unsupervised_train.py <span style='color:#111;'> 30.31KB </span>","children":null,"spread":false},{"title":"models.py <span style='color:#111;'> 25.68KB </span>","children":null,"spread":false},{"title":"unsupervised_train__.py <span style='color:#111;'> 16.72KB </span>","children":null,"spread":false},{"title":"utils.py <span style='color:#111;'> 3.67KB </span>","children":null,"spread":false},{"title":"minibatch.py <span style='color:#111;'> 12.75KB </span>","children":null,"spread":false},{"title":"metrics.py <span style='color:#111;'> 1.45KB </span>","children":null,"spread":false},{"title":"neigh_samplers_.py <span style='color:#111;'> 26.26KB </span>","children":null,"spread":false},{"title":"supervised_train_.py <span style='color:#111;'> 42.60KB </span>","children":null,"spread":false},{"title":"supervised_models.py <span style='color:#111;'> 8.34KB </span>","children":null,"spread":false},{"title":"neigh_samplers_org.py <span style='color:#111;'> 818B </span>","children":null,"spread":false},{"title":"supervised_test.py <span style='color:#111;'> 12.11KB </span>","children":null,"spread":false},{"title":"__init__.py <span style='color:#111;'> 70B </span>","children":null,"spread":false},{"title":"aggregators.py <span style='color:#111;'> 23.10KB </span>","children":null,"spread":false},{"title":"layers.py <span style='color:#111;'> 3.69KB </span>","children":null,"spread":false},{"title":"inits.py <span style='color:#111;'> 934B </span>","children":null,"spread":false},{"title":"prediction.py <span style='color:#111;'> 5.26KB </span>","children":null,"spread":false},{"title":"supervised_train.py <span style='color:#111;'> 33.93KB </span>","children":null,"spread":false}],"spread":false}],"spread":false}],"spread":true}]

评论信息

免责申明

【只为小站】的资源来自网友分享,仅供学习研究,请务必在下载后24小时内给予删除,不得用于其他任何用途,否则后果自负。基于互联网的特殊性,【只为小站】 无法对用户传输的作品、信息、内容的权属或合法性、合规性、真实性、科学性、完整权、有效性等进行实质审查;无论 【只为小站】 经营者是否已进行审查,用户均应自行承担因其传输的作品、信息、内容而可能或已经产生的侵权或权属纠纷等法律责任。
本站所有资源不代表本站的观点或立场,基于网友分享,根据中国法律《信息网络传播权保护条例》第二十二条之规定,若资源存在侵权或相关问题请联系本站客服人员,zhiweidada#qq.com,请把#换成@,本站将给予最大的支持与配合,做到及时反馈和处理。关于更多版权及免责申明参见 版权及免责申明