N Neurarch Architectures Models Checks Data Docs Open the app

Comparisons / LightGCN vs GraphSAGE Recommender

LightGCN vs GraphSAGE Recommender

A stripped graph convolution against a sampled one.

GraphSAGE Recommender has 70M fewer parameters than LightGCN: 8 layers added, 12 removed, 1 changed.

Baseline

LightGCN

Layers
11
Parameters
70M
Input
1
Output
64
Forward-passes
yes
Est. train cost
$0.065
Compared

GraphSAGE Recommender

Layers
8
Parameters
263K
Input
128
Output
128 × 64
Forward-passes
yes
Est. train cost
$0.042

The deltas

Every number is GraphSAGE Recommender relative to LightGCN.

Parameters
-70M (-99.6%)
Layers
-3
Added
8
Removed
12
Changed
1
Unchanged
1

Which GPUs each one fits

Each side is measured at its own declared input (1 against 128). Both columns are right about their own model; the difference between them is not a fact about the designs.

GPULightGCNGraphSAGE Recommender
T4 16GBfitsfits
A100 40GBfitsfits
H100 80GBfitsfits

Layer by layer

Aligned in topological order. 1 of 22 rows are the same layer with the same parameters.

Hide all 22 rows
LightGCNGraphSAGE Recommender
LayerParamsOutputLayerParamsOutput
1removedUser ID
Input
1
2changed
shape
Item ID
Input
1Node Features
Input
128
3removedE_user(0)
Embedding
6.4M1 × 64
4removedE_item(0)
Embedding
64M1 × 64
5removedGraphConv(1)
Graph Conv
4.2K1 × 64
6removedGraphConv(1)
Graph Conv
4.2K1 × 64
7removedGraphConv(2)
Graph Conv
4.2K1 × 64
8removedGraphConv(2)
Graph Conv
4.2K1 × 64
9removedGraphConv(3)
Graph Conv
4.2K1 × 64
10removedGraphConv(3)
Graph Conv
4.2K1 × 64
11removedLayer Combine (mean)
Mean
64
12removedLayer Combine (mean)
Mean
64
13removedDot Score
Matmul
64
14addedSAGE Layer 1
Graph SAGE
66K128 × 64
15addedReLU
Relu
128 × 64
16addedDropout
Dropout
128 × 64
17addedSAGE Layer 2
Graph SAGE
131K128 × 64
18addedReLU
Relu
128 × 64
19addedDropout
Dropout
128 × 64
20addedSAGE Layer 3
Graph SAGE
66K128 × 64
21addedLayerNorm
Layer Norm
256128 × 64
22sameScore
Output
64Node Embedding
Output
128 × 64

What this is not

Take it further

Open either graph in the editor, change it, and check it again: LightGCN · GraphSAGE Recommender

Compare any two models of your own, including anything on Hugging Face: the comparison tool.

Machine-readable: this page as markdown · the pair index · POST https://www.neurarch.com/api/v1/plan for a graph of your own.