euler icon indicating copy to clipboard operation
euler copied to clipboard

A distributed graph deep learning framework.

Results 104 euler issues
Sort by recently updated
recently updated
newest added

tf_euler.initialize_graph({'mode':'remote','zk_server':'9.22.56.116:2182','zk_path':'/euler','shard_num':2,'num_retries':1}) 2020-08-20 08:47:57.353119: W /Euler-OpenSource/euler/common/zk_server_monitor.cc:256] Online node: 0#9.22.56.116:20057. 2020-08-20 08:47:57.353209: W /Euler-OpenSource/euler/common/zk_server_monitor.cc:256] Online node: 1#9.22.56.102:21695. 2020-08-20 08:47:57.358419: W /Euler-OpenSource/euler/client/client_manager.h:87] shard number: 2 2020-08-20 08:47:57.358676: F /Euler-OpenSource/euler/client/client_manager.cc:33] Invalid weight meta failed,...

看公众号发的视频交流群过期了,我就来看看issue回复是不是真的快

[ 13%] Performing build step for 'jemalloc' cp include/jemalloc/internal/private_namespace.gen.h include/jemalloc/internal/private_namespace.gen.h [ 13%] Performing install step for 'jemalloc' Skipping install step. [ 13%] Completed 'jemalloc' [ 13%] Built target jemalloc

运行bash run_ppi.sh时报错: /usr/bin/python: No module named tf_euler.__main__; 'tf_euler' is a package and cannot be directly executed 请问如何解决?谢谢

Hi, Thanks for developing this project, I also noticed that alibaba open-sourced [graph-learn](https://github.com/alibaba/graph-learn) recently, can anyone point out the differences between this two projects ?

如题,有没有提供采样一个节点的非一阶邻居节点的方法呀,用来做无监督的负样本

我用自己写的模型,在分布式下跑,10个分片,9千万结点,启用2个ps会oom,但是启用3个ps时第三个启用了但是没有数据进入。 测试了用line去跑一份简单的数据,3个ps1个worker,只有第一个ps工作,第23个ps没有参与计算,该如何解决?

因为单个模型训练会占用大量资源,是否有资源管理/资源隔离的机制?比如结合yarn?

@alinamimi @renyi533 请教3个PS时,model_dir 是不是只能指定hdfs目录呀,我指定了本地的同名目录,结果有错误,指定hdfs目录需要配置什么呀,目录格式是什么呢? 感谢解答!

我用了examples目录下的graphsage,命令如下: ``` python run_graphsage.py --dataset reddit --run_mode train --batch_size 500 --learning_rate 0.01 ``` 运行结果如下: ![image](https://user-images.githubusercontent.com/1694657/92110529-70abf000-ee1d-11ea-8254-74520ce5b2e5.png) 请教一下各位,这个问题应该如何解决。十分感谢!