Closed MrKing8810 closed 6 years ago
Pre-trained node embeddings can be obtained by any other network embedding method. Of course you can randomly initialize the node embeddings, but it may take you more time to fine-tune the hyper-parameters to observe a good result. In other words, a relatively "good" set of initial node embeddings can help GAN to converge better.
@hwwang55 Thank you for explaining the reason in detail
@hwwang55 谢谢啦~!
@hwwang55 Could you please tell me that what method did you use to obtain the pre-trained node embeddings.
@fukien It's DeepWalk
@hwwang55 Noted, thanks.
Me too, I would like to ask for the same question. Can I use random number to initialize the pre_trained embedding file