chenxing1999/recsys-benchmark

Benchmark source code for multiple recommendation backbone

RecSys Benchmark

Installation

# clone this repo
git clone [email protected]:chenxing1999/recsys-benchmark.git

# cd to folder
cd recsys-benchmark

# Create virtual env
python -m venv env
source env/bin/activate

# install
pip install -e '.[dev]'

Quickstart

Train SGL-WA:

python scripts/lightgcn/train_lightgcn.py configs/yelp2018/base_config.yaml

Run Hyperparam search with Optuna

mkdir -p checkpoints/lightgcn/sgl-wa.pth
python scripts/lightgcn/exp_find_hparams.py -c configs/yelp2018/base_config.yaml -l logs/sgl-wa -p checkpoints/lightgcn/sgl-wa.pth

For more detail with LightGCN, click here

For more detail with DeepFM, click here. DCN-Mix uses the same API with DeepFM.

Other artifacts

TBD

Acknowledge

This source code is based on:

https://github.com/gusye1234/LightGCN-PyTorch
https://github.com/rixwew/pytorch-fm

recsys-benchmark's Issues

Add weight decay to PEP