search

tkipf / gcn

Implementation of Graph Convolutional Networks in TensorFlow
MIT License
7.08k stars 2k forks source link

difference between x and allx #101

Open amrmohammed123 opened 5 years ago

amrmohammed123 commented 5 years ago

what is the use of x in load_data function as from what I see it's just loaded and isn't used, but the one used is allx, what are the differences between them.

tkipf commented 5 years ago

If I remember correctly, x was only used for inductive semi-supervised learning tasks in the paper that released that benchmark (see README), and does not contain all node feature vectors. For transductive learning (the setting considered in this work), simply use allx which contains all feature vectors.

On Fri, Apr 5, 2019 at 12:56 PM amrmohammed123 notifications@github.com wrote:

what is the use of x in load_data function as from what I see it's just loaded and isn't used, but the one used is allx, what are the differences between them.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/tkipf/gcn/issues/101, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYDqcC7QKXsaiNiDsSF1upSkTYBISks5vdyvAgaJpZM4ceqxI .

b224618 commented 4 years ago

If I remember correctly, x was only used for inductive semi-supervised learning tasks in the paper that released that benchmark (see README), and does not contain all node feature vectors. For transductive learning (the setting considered in this work), simply use allx which contains all feature vectors. … On Fri, Apr 5, 2019 at 12:56 PM amrmohammed123 @.***> wrote: what is the use of x in load_data function as from what I see it's just loaded and isn't used, but the one used is allx, what are the differences between them. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#101>, or mute the thread https://github.com/notifications/unsubscribe-auth/AHAcYDqcC7QKXsaiNiDsSF1upSkTYBISks5vdyvAgaJpZM4ceqxI .

Actually allx does not contain all features. There are 2708 nodes in cora. I tried loading the data and found the shape of allx is 1708x1433, containing features of 1708 nodes except the 1000 test nodes