How to support graph batch-norm? #10
Comments
|
Can we get the node states and apply batch norm outside the graph?
…On Tue, Jun 19, 2018, 10:29 PM GaiYu0 ***@***.***> wrote:
With current API, we can implement batch-norm by adding a shadow node and
pulling/pushing globally. However, this may not be very user-friendly. An
ideal syntax is
batch_norm(node_reprs['x'])
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
<#10>, or mute the thread
<https://github.com/notifications/unsubscribe-auth/AD3qZR-CTAnLNfK_TwmvWUTGEiM0epaGks5t-Py7gaJpZM4UtgtH>
.
|
|
Yes, we can always bypass our node-centric API to do batch-norm. |
6 tasks
mufeili
added a commit
that referenced
this issue
Jan 25, 2022
* Update * Update * Fix * Update * Update * Update * Update * Fix * Update * Update * Update * Update * Fix lint * lint * Update * Update * lint fix * Fix CI * Fix * Fix CI * Update * Fix * Update * Update * Augmentation (#10) * Update * PPR * Update * Update * Update * Update * Update * Update * Update * Update * Update * Update * Update * Update * CI * lint * lint * Update * Update * Fix AddEdge * try import * Update * Fix * CI Co-authored-by: Ubuntu <ubuntu@ip-172-31-31-136.us-west-2.compute.internal> Co-authored-by: Minjie Wang <wmjlyjemaine@gmail.com>
GMNGeoffrey
added a commit
to GMNGeoffrey/dgl
that referenced
this issue
Jan 29, 2025
These changes are on the original DGL source code. These plus running `script/hipify-inplace.sh` yields the hipified version of DGL which is identical to the code currently in nod-ai#9. The version in this commit should still run with CUDA. This obviously shouldn't be merged into the same branch as PRs dmlc#1 through dmlc#9. The idea is that this would be the PR we would need for upstream (although I'm guessing they would actually want it in smaller chunks).
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
With current API, we can implement batch-norm by adding a shadow node and pulling/pushing globally. However, this may not be very user-friendly. An ideal syntax is
batch_norm(node_reprs['x'])The text was updated successfully, but these errors were encountered: