the normalize function in utils.py

[wangxiaoyunanne]

Hi,
The normalize function in utils.py only normalize the row of adjacency matrix. While for the tensorflow version the implementation is different. You normalized both row and column. I am wondering will this lead to a difference for performance of GCN in accuracy ?

Best,
Xiaoyun

[tkipf]

Both versions result in similar performance but slight differences in classification accuracy are possible.

…

On Mon 2. Jul 2018 at 23:56 wangxiaoyunanne ***@***.***> wrote: Hi, The normalize function in utils.py only normalize the row of adjacency matrix. While for the tensorflow version the implementation is different. You normalized both row and column. I am wondering will this lead to a difference for performance of GCN in accuracy ? Best, Xiaoyun — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#11>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYLts2vOdZh-TXRQ4Hi9BUajEoEu0ks5uCqU0gaJpZM4VADWS> .

[mathfinder]

Hi,
I want to know which version is better in the classification.

[tkipf]

Hard to say, it’s often best to try both variants.

…

On Sun 2. Dec 2018 at 22:30 Defa Zhu ***@***.***> wrote: Hi, I want to know which version is better in the classification. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYLg2pZ95Tn5xJfEL--SkcGyAXJ2kks5u1JrkgaJpZM4VADWS> .

[AllenWu18]

Well, I think in the function normalize(mx) it normalizes the features by row,
but if the node features are not one-hot form, and each node has its own feature
like node1 : 1,3,7,20,57, which each feature represents different information( just like something similar in traditional machine learning, the different features has different meaning, like age, height, etc)
the each column of features has different dimension,
should I normalize them by column, instead of row??

[tkipf]

Yes, you could normalize each feature to zero mean, unit standard deviation per column.

…

On Tue 16. Apr 2019 at 06:00 AllenWu18 ***@***.***> wrote: Well, I think in the function normalize(mx) it normalizes the features by row, but if the node features are not one-hot form, and each node has its own feature like node1 : 1,3,7,20,57, which each feature represents different information( just like something similar in traditional machine learning, the different features has different meaning, like age, height, etc) the each column of features has different dimension, should I normalize them by column, instead of row?? — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#11 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYNzE90AZurjH82ikHzwfokAHGtJ3ks5vhUrEgaJpZM4VADWS> .

[AllenWu18]

THX,I'll have a try