Can GNNexplainer node importance be compared column wise that is can we see how single node importance change as output changes #9224
-
|
I have created a csv file containing node importance of all my test samples now I want to know if I can take a single column i.e. importance of a single node accross multiple column. I want to see how the importance vary as my output changes is this a valid attempt? |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments 1 reply
-
|
Can you clarify? Do you want to change the input and see how the importance varies? I think fidelity or unfaithfulness metrics are what you are looking for, see https://pytorch-geometric.readthedocs.io/en/latest/modules/explain.html#explanation-metrics. |
Beta Was this translation helpful? Give feedback.
-
|
No I am more interested to know that correlation between ground-truth and
and node importance. I want to know if this is a valid step to take ? can I
take the correlation between not importance and GroundTruth and make
implications from their relationship?
…On Mon, 22 Apr, 2024 at 6:08 PM, Matthias Fey ***@***.***> wrote:
Can you clarify? Do you want to change the input and see how the
importance varies? I think fidelity or unfaithfulness metrics are what you
are looking for, see
https://pytorch-geometric.readthedocs.io/en/latest/modules/explain.html#explanation-metrics
.
—
Reply to this email directly, view it on GitHub
<#9224 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/ARAQLCIBNEERBU2GOXWYB7TY6T4SJAVCNFSM6AAAAABGR3GFBWVHI2DSMVQWIX3LMV43SRDJONRXK43TNFXW4Q3PNVWWK3TUHM4TCOBYGAZDG>
.
You are receiving this because you authored the thread.Message ID:
***@***.***
com>
|
Beta Was this translation helpful? Give feedback.
-
|
Yes, I think that's definitely possible. |
Beta Was this translation helpful? Give feedback.
Yes, I think that's definitely possible.