Trying to set up network for regression task

[emanuelazcona]

Could you possibly point me on where to start with this so I can set this up for a regression task? Plan on using L2_loss.

I have multiple samples of data that lie on the same graph, however the features at each node is a vector. I was wondering how to get around since I'm dealing with a 3D tensor:

X : (Samples x Nodes x features)
Y: (Samples x Nodes x features) as well

Most of the operations in both implementations seem to only work for X being a matrix, unless I'm mistaken.

Of course, my adjacency & laplacian will remain 2D matrices for all operations (there the Cheby. will be too)

[tkipf]

For regression, simply swap out the loss function (L2 instead of cross entropy should do fine). If you have multiple separate graph instances, just concatenate their feature matrices. You can accordingly ‘concatenate’ the adjacency matrices into a large N_total x N_total matrix with block-diagonal structure. N_total is the total number of nodes after concatenation.

…

On Tue 17. Jul 2018 at 05:03 Emanuel Azcona ***@***.***> wrote: Could you possibly point me on where to start with this so I can set this up for a regression task? Plan on using L2_loss. I have multiple samples of data that lie on the same graph, however the features at each node is a vector. I was wondering how to get around since I'm dealing with a 3D tensor: X : (Samples x Nodes x features) Y: (Samples x Nodes x features) as well Most of the operations in both implementations seem to only work for X being a matrix, unless I'm mistaken. Of course, my adjacency & laplacian will remain 2D matrices for all operations (there the Cheby. will be too) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#34>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYFG-G4dp7UcqfKDnXoZBKhHcbEr4ks5uHWIWgaJpZM4VSJgi> .

[emanuelazcona]

Thanks for the tip! Will try now.

Also, have you guys worked on pooling for this? I know Defferrard et al. built their Chebynet with pooling. Just wondering if there is a Keras implementation yet.

[tkipf]

Have a look here at my previous works: https://scholar.google.com/citations?user=83HL5FwAAAAJ Aside from global gated pooling (in the recent MolGAN paper) and some form of self-attentive pooling (in the NRI paper), we haven’t done much along these lines.

…

On Wed 18. Jul 2018 at 22:18 Emanuel Azcona ***@***.***> wrote: Thanks for the tip! Will try now. Also, have you guys worked on pooling for this? I know Defferrard et al. built their's Chebynet with pooling. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#34 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHAcYCx0fA7x7JJE_kqMqlNR0sNhJ3DGks5uH6YqgaJpZM4VSJgi> .