compare CASC to GraphSAGE on the experiments they performed in their papers

[jovo]

https://arxiv.org/pdf/1706.02216.pdf

[bdpedigo]

@jovo ndd appropriate?

[jovo]

Seems like it to me? JL has graphsage working I understand.

On Wed, Oct 14, 2020 at 11:54 AM Benjamin Pedigo [email protected] wrote:

•  External Email - Use Caution      *
  

@jovo https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fjovo&data=02%7C01%7Cjovo%40jhu.edu%7C2bfa7ba357ac4a67836708d870596638%7C9fa4f438b1e6473b803f86f8aedf0dec%7C0%7C0%7C637382876548704559&sdata=4HXiXJ0uYXzuJervXDmdrKOn6r0znVZ2jMw7U3dAO7A%3D&reserved=0 ndd appropriate?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fmicrosoft%2Fgraspologic%2Fissues%2F536%23issuecomment-708495261&data=02%7C01%7Cjovo%40jhu.edu%7C2bfa7ba357ac4a67836708d870596638%7C9fa4f438b1e6473b803f86f8aedf0dec%7C0%7C0%7C637382876548704559&sdata=dmwkuHglae6eS5zYndthL3MhI8SypBtb3FXfqHl%2Bwzw%3D&reserved=0, or unsubscribe https://nam02.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAAAKG4RS2PL3FC2RMUUAUKLSKXCSFANCNFSM4SQYCUCQ&data=02%7C01%7Cjovo%40jhu.edu%7C2bfa7ba357ac4a67836708d870596638%7C9fa4f438b1e6473b803f86f8aedf0dec%7C0%7C0%7C637382876548714549&sdata=fd9jU68w1cmLGMkEocb2wfKxLtragR1FRszmIYDcuD8%3D&reserved=0 .

-- Typed with thumbs

[spencer-loggia]

Possibly interested in this issue for sprint 2

[loftusa]

also interested in this for sprint 2

[bdpedigo]

@JonathanLarson do you have an implementation of GraphSage you like/use?

[zeou1]

I tried running the PPI dataset with the GraphSAGE GitHub code by first running the unsupervised model and then the PPI logistic classificer, but it just outputs the F1 scores for each node in the test graph. The GitHub code does not seem immediately ready to be used to run the experiments as detailed in their paper, but I believe I can write a new Jupyter Notebook to run their model and then classify with my own code instead of theirs.

[zeou1]

Definition of Done (First Draft):

1. Write a Jupyter Notebook for the experiment to replicate the PPI dataset result in Table 1 of the paper
2. Do the same thing as step 1 but with CASC

I feel like the two steps above should be achievable within the time constraint of the NDD class, but I do welcome any feedback on this part.

[zeou1]

Are there other ways to compare the results of the experiments other than F1 scores? Or is the F1 score what we are looking for?