MTandHJ/LT-OCF

LT-OCF: Learnable-Time ODE-based Collaborative Filtering, CIKM'21

LT-OCF: Learnable-Time ODE-based Collaborative Filtering

Introduction

This is the repository of our accepted CIKM 2021 paper "LT-OCF: Learnable-Time ODE-based Collaborative Filtering". Paper is available on arxiv

Citation

Please cite our paper if using this code.

@inproceedings{choi2021ltocf,
  title={LT-OCF: Learnable-Time ODE-based Collaborative Filtering},
  author={Choi, Jeongwhan and Jeon, Jinsung and Park, Noseong},
  booktitle={Proceedings of the 30th ACM International Conference on Information and Knowledge Management},
  year={2021},
  organization={ACM}
}

Our proposed LT-OCF

Our proposed dual co-evolving ODE

---

Setup Python environment for LT-OCF

Install python environment

conda env create -f environment.yml   

Activate environment

conda activate lt-ocf

---

Reproducibility

Usage

In terminal

• Run the shell file (at the root of the project)

# run lt-ocf (gowalla dataset, rk4 solver, learnable time)
sh ltocf_gowalla_rk4.sh

# run lt-ocf (gowalla dataset, rk4 solver, fixed time)
sh ltocf_gowalla_rk4_fixed.sh

Arguments (see more arguments in parse.py)

• gpuid
  • default: 0
• dataset
  • gowalla, yelp2018, amazon-book
• model
  • ltocf
• solver
  • euler, rk4, implicit_adams, dopri5
• adjoint
  • False, True
• K
  • 1, 2, 3, 4
• learnable_time
  • True, False
• dual_res
  • False, True

Usage

conda activate lt-ocf

hyper-parameters:

• decay: [1.e-4, 1.e-3, 1.e-2]
• K: [2, 3, 4]
• lr: [1.e-5, 1.e-4, 1.e-3, 1.e-2]

Gowalla_m1

python main.py --dataset="Gowalla_m1" --model="ltocf" --solver="rk4" --adjoint=False --K=4 --learnable_time=True --dual_res=False --lr=1e-4 --lr_time=1e-6 --decay=1e-4 --topks="[20]" --comment="learnable_time" --tensorboard=1 --gpuid=3

MovieLens1M_m2

python main.py --dataset="MovieLens1M_m2" --model="ltocf" --solver="rk4" --adjoint=False --K=4 --learnable_time=True --dual_res=False --lr=1e-3 --lr_time=1e-6 --decay=1e-4 --topks="[20]" --comment="learnable_time" --gpuid=0

• {'precision': array([0.14490203]), 'recall': array([0.252613]), 'ndcg': array([0.24009889])}

AmazonCDs_m1

python main.py --dataset="AmazonCDs_m1" --model="ltocf" --solver="rk4" --adjoint=False --K=4 --learnable_time=True --dual_res=False --lr=1e-3 --lr_time=1e-6 --decay=1e-4 --topks="[20]" --comment="learnable_time" --gpuid=1

• {'precision': array([0.0273657]), 'recall': array([0.15701499]), 'ndcg': array([0.09929412])}

AmazonElectronics_m1

python main.py --dataset="AmazonElectronics_m1" --model="ltocf" --solver="rk4" --adjoint=False --K=4 --learnable_time=True --dual_res=False --lr=1e-4 --lr_time=1e-6 --decay=1e-3 --topks="[20]" --comment="learnable_time" --gpuid=2

• {'precision': array([0.01756098]), 'recall': array([0.12778234]), 'ndcg': array([0.07546926])}