The code does not support transfer learning across graphs currently

[PonderLY]

Hi GentleZhu,

I found that the code does not handle the scenario when data-id and data-src are different. I am not sure whether I miss any key part in the code.

Thanks and Regards,
PonderLY

[Skyorca]

EGI is a two-stage model, thus you can dump the best embedding model and the best classifier seperately while training on source graph, then you can load target graph and apply the dumpled models to it :)

[abdullahgarra]

Following on the previous question, if i remove the if else block here and always load the best model it should also perform the transferability test with the correct parameters passed.
for example:
for training on Europe:
python3 run_airport.py --file-path=data/europe-airports.edgelist --label-path=data/labels-europe-airports.txt --n-dgi-epochs=100 --n-hidden=32 --self-loop --gpu=-1 --n-layers=1 --dgi-lr=0.01 --model-id=2 --model-type=2

and then for testing on USA you'd run: (you zero out the n-dgi-epochs )
python3 run_airport.py --file-path=data/usa-airports.edgelist --label-path=data/labels-usa-airports.txt --data-id=usa --data-src=europe --model-type=2 --model-id=2 --n-dgi-epochs=0 --n-classifier-epochs=100 --n-hidden=32 --self-loop --gpu=-1