TD3: Tucker Decomposition Based Dataset Distillation Method for Sequential Recommendation

Contents

This repository utilizes PyTorch and modern experiment manager tools, Hydra and Wandb.

Datasets are downloaded with Rebole, and preprocessed with [link]

Directory structure:

.
├── config
│   ├── bert4rec.yaml
│   ├── default.yaml
│   ├── gru4rec.yaml
│   ├── narm.yaml
│   └── sasrec.yaml
├── data
│   ├── preprocess.ipynb
│   ├── processed
│   │   ├── epinions
│   │   │   └── epinions.inter
│   │   ├── magazine
│   │   │   └── magazine.inter
│   │   ├── ml-100k
│   │   │   └── ml-100k.inter
│   │   └── ml-1m
│   │       └── ml-1m.inter
│   └── raw
├── environment.yml
├── README.md
├── script
│   ├── baseline.sh
│   ├── epinions.sh
│   ├── magazine.sh
│   ├── ml100k.sh
│   ├── ml1m.sh
│   └── test.sh
└── src
    ├── baseline.py
    ├── data.py
    ├── distilled_data.py
    ├── evaluator.py
    ├── main.py
    ├── model.py
    ├── pretrainer.py
    ├── trainer.py
    └── utils.py

Run Scripts

1. Clone this repository.

   $ git clone https://github.com/Maitouer/TD3.git
   $ cd TD3
   

2. Prepare environment for Python 3.10 and install requirements.

   $ conda env export > environment.yml
   

3. Run experiments.

   $ ./script/magazine.sh
   $ ./script/epinions.sh
   $ ./script/ml100k.sh
   $ ./script/ml1m.sh