graphsage train_unsupervised.py get killed while save embeddings about graph-learn

[archwalker]

Can you please take a look at the memory usage? It is generally caused by memory consumption.

[kevinyuan]

The process is occupying 12g memory, and when the embedding is being written, it tried to allocate more memory and the peak system memory utilization is about 95%. Thanks. Kevin Yuan 在 2020年4月28日，下午1:10，archwalker <[email protected]> 写道：  Can you please take a look at the memory usage? It is generally caused by memory consumption. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#39 (comment)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AABYOHKWY4RHL6VAM27JOVDROZQMXANCNFSM4MSCZTPA>.

[archwalker]

Yes, that relates to the memory problem. We use np.save to save the whole embedding matrix at once, which would cause a huge memory consumption problem, but this problem does not show up when we test the program on a server. I will fix this problem.

For now, you may:

1. try to save with pickle, it might work.
2. separate the embedding matrix and save them one by one.

[archwalker]

Hi, while modifying the code to save embeddings partially, we found a memory leak problem also related to your case and solve it #46 , you can try again to see whether it has been fixed.