alibaba / alibaba/euler

基于同样的ckpt多次导出的embedding表都不一样?

Open
#104 4 comments 0 reactions 0 assignees View on GitHub
Dominant language
C++
Stars
2.9k
Forks
553
PR merge metrics
No merged PRs in 30d

Description

运行官网的ppi数据集,跑的是graphsage_supervised算法模型,训练后,然后进行embedding表的导出,
(1) 第1次执行命令如下:
python -m tf_euler --model_dir ./model_graphsage_supervised_local --max_id 56944 --feature_idx 1 --feature_dim 50 --label_idx 0 --label_dim 121 --model graphsage_supervised --mode lsave_embedding --data_dir ppi

导出的embedding大小如下:
58311808 embedding.npy
然后将embedding.npy转换为embedding表,其大小为:
163233692 embedding_npy_new_3
得到第一行的值为:
1 0.16256002 0.6374952 0.6620392 1.0137732 -0.43433663 -0.21821757 -0.074182674 ......
(2) 再次执行如下命令:
python -m tf_euler --model_dir ./model_graphsage_supervised_local --max_id 56944 --feature_idx 1 --feature_dim 50 --label_idx 0 --label_dim 121 --model graphsage_supervised --mode save_embedding --data_dir ppi

导出的embedding大小如下:
58311808 embedding.npy
然后将embedding.npy转换为embedding表,其大小为:
163225039 embedding_npy_new_4
得到第一行的值为:
1 0.116333544 0.6290663 0.47425652 0.98439133 -0.5063977 -0.24195752 -0.09124704......

@yangsiran 从(1)和(2)看到,基于同样的训练模型ckpt基础上进行embedding表导出,先后得到的embedding表是不同的,查看各表的第1行对应的权重参数是不同的,所以,想问一下,多次基于相同的ckpt导出的embedding表对应的权重参数怎么都不一样?

Contributor guide

No contributing guide indexed for this repository

Research direction

Reproduce the PPI GraphSAGE supervised export from model_graphsage_supervised_local with the two commands shown, comparing mode lsave_embedding and mode save_embedding. Inspect the generated embedding.npy files and their first rows; done means determining why the same checkpoint produces different embeddings and documenting or correcting the behavior.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.