dmlc / dgl

Python package built to ease deep learning on graph, on top of existing DL frameworks.
http://dgl.ai
Apache License 2.0
13.23k stars 2.99k forks source link

[GraphBolt] Refactor sampling #7367

Closed RamonZhou closed 2 months ago

RamonZhou commented 2 months ago

Description

Refactor NeighborSampling and TemporalNeighborSampling.

Checklist

Please feel free to remove inapplicable items for your PR.

Changes

dgl-bot commented 2 months ago

To trigger regression tests:

dgl-bot commented 2 months ago

Commit ID: d16ed9cdea6b738ef2ad517c1e147780d4faa0fa

Build ID: 1

Status: ✅ CI test succeeded.

Report path: link

Full logs path: link

mfbalin commented 2 months ago

How did the MSVC bug get resolved?

mfbalin commented 2 months ago

Depends on #7369.

mfbalin commented 2 months ago

@RamonZhou can you merge latest changes by updating branch and test the hetero example?

dgl-bot commented 2 months ago

Commit ID: 8a880c28044cb65bc2f4d20990a764b5936f1e00

Build ID: 2

Status: ✅ CI test succeeded.

Report path: link

Full logs path: link

RamonZhou commented 2 months ago

@RamonZhou can you merge latest changes by updating branch and test the hetero example?

@mfbalin I tested hetero_rgcn (CPU) on my local machine and it goes from 248.9s to 235.1s (per epoch), 5.5% improvement.

mfbalin commented 2 months ago

@RamonZhou can you merge latest changes by updating branch and test the hetero example?

@mfbalin I tested hetero_rgcn (CPU) on my local machine and it goes from 248.9s to 235.1s (per epoch), 5.5% improvement.

That is because no GPU is there for forward backward. More improvement is expected when forward backward runs on the GPU. Cool, good work overall!