This is the official implementation of Graph-enhanced Optimizers for Structure-aware Recommendation Embedding Evolution.

Note: We provide this code for precise reproduction of the results reported in the paper, which may be a bit troublesome to prepare the environment. SEvo-latest provides an alternative supporting latest PyTorch/PyG/FreeRec.

Requirements

Run the following commands to prepare the environment (CUDA: 11.3/11.6):

conda create --name=PyT12 python=3.9; conda activate PyT12; bash setup_env.sh

Usage

We provide configs and experimental logs for the Neumann series approximation with re-scaling. You can re-run them and try some other hyperparameters:

python main.py --config=configs/xxx.yaml --optimizer=AdamWSEvo --aggr=neumann --L=3 --beta3=0.99 --H=1

• optimizer: AdamWSEvo|AdamW|AdamSEvo|Adam|SGDSEvo|SGD
• aggr: neumann|iterative
• L: the number of layers for approximation
• beta3: $\beta$
• H: The maximum walk length allowing for a pair of neighbors