search

DeepGraphLearning / graphvite

GraphVite: A General and High-performance Graph Embedding System
https://graphvite.io
Apache License 2.0
1.22k stars 151 forks source link

node2vec implementation "Segmentation fault" #18

Closed PJthunder closed 4 years ago

PJthunder commented 5 years ago

My machine has 32core 200 GB memory + 4 Tesla V100. I tried the node2vec config yaml. It reports "segmentation fault" before the training begins.

The same thing appears when I try to use node2vec to train on another dataset. Is there any bug in the node2vec implementation.

p.s. I choose "auto" for the parameter "num_partition".

KiddoZhu commented 5 years ago

It's node2vec's fault. Node2vec uses 2nd order random walk and is less scalable than DeepWalk / LINE in memory cost.

In benchmarks, you can find that node2vec can only runs up to 5 million edges.

PJthunder commented 5 years ago

So graphvite version of node2vec cannot handle the youtube network, which has 5M edges? Since I try to run the youtube yaml for node2vec but still fails, wondering whether it is due to some bugs in the partition code or it's model capacity.

KiddoZhu commented 5 years ago

Sorry, I found the same issue on Youtube. I will check that.

KiddoZhu commented 5 years ago

Fixed. Will update it to conda.

KiddoZhu commented 4 years ago

Updated to conda.

porimol commented 4 years ago

@KiddoZhu

I'm getting the same error "Segmentation fault"!