Closed PJthunder closed 4 years ago
Yes, link prediction is supported for node embedding, although there isn't an example.
You can find the interface here, in GraphApplication.link_prediction
. The evaluation file format is
head tail label
, where positive label is 1 and negative label is 0.
If you can't guarantee that evaluation data is excluded in the training set, you can specify the training edge list as filter_file
.
For example, a config might be
evaluate:
task: link prediction
file_name: test.txt
filter_file: train.txt
Added in v0.2.0.
I have read over some config files. It looks like all the evaluation for node embedding is node classification. Is that possible to run link prediction task on node embedding tasks? If so, what's the format of the evaluation file and the config files?