freerec 0.6.3

PyTorch library for recommender systems

FreeRec is a personal repository designed for easy data pre-processing and model training. I am a beginner in the field of recommender systems, so much of FreeRec's designs may not be as effective. In addition, you are free to specify your own framework based on FreeRec.

Requirements:

Python == 3.9 | PyTorch == 1.12.1 | TorchData == 0.4.1 | PyG (optional)

conda create --name=PyT12 python=3.9
conda activate PyT12

GPU

pip install torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install torchdata==0.4.1

• Linux

pip install torch_geometric==2.1.0.post1
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcu116/torch_scatter-2.0.9-cp39-cp39-linux_x86_64.whl
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcu116/torch_sparse-0.6.15%2Bpt112cu116-cp39-cp39-linux_x86_64.whl

• Windows

pip install torch_geometric==2.1.0.post1
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcu116/torch_scatter-2.0.9-cp39-cp39-win_amd64.whl
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcu116/torch_sparse-0.6.15%2Bpt112cu116-cp39-cp39-win_amd64.whl

CPU

conda install pytorch==1.12.1 cpuonly -c pytorch
pip install torchdata==0.4.1

• Linux

pip install torch_geometric==2.1.0.post1
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcpu/torch_scatter-2.0.9-cp39-cp39-linux_x86_64.whl
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcpu/torch_sparse-0.6.15%2Bpt112cpu-cp39-cp39-linux_x86_64.whl

• Windows

pip install torch_geometric==2.1.0.post1
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcpu/torch_scatter-2.0.9-cp39-cp39-win_amd64.whl
pip install https://data.pyg.org/whl/torch-1.12.0%2Bcpu/torch_sparse-0.6.15%2Bpt112cpu-cp39-cp39-win_amd64.whl

Installation

pip install freerec

or (for latest)

pip install git+https://github.com/MTandHJ/freerec.git

Data Pipeline

Note: To make dataset, please download corresponding Atomic files from [RecBole]. Then, run `make_dataset.ipynb'.

Training Flow

Reference Code

• TorchRec: https://github.com/pytorch/torchrec
• DeepCTR-Torch: https://github.com/shenweichen/DeepCTR-Torch
• FuxiCTR: https://github.com/xue-pai/FuxiCTR
• BARS: https://github.com/openbenchmark/BARS
• RecBole: https://github.com/RUCAIBox/RecBole

Acknowledgements

Thanks to ChatGPT for the annotation of some code. For this reason, some of the comments may be illogical.