PyTorch and PyTorch geometric based implementation of Graph Convolutional Matrix Completion.
This repository is NOT an official implementation of that paper.
The official implementation is this (Tensorflow).
Our experimental result is shown below and it doesn't reach to the score of the original paper.
- Setup a virtual environment of python 3.6 or newer
- Install requirements (pip)
pip install -r requirements/1.txt pip install --verbose --no-cache-dir -r requirements/2.txt pip install -r requirements/3.txt
Installation of Pytorch Geometric is very troublesome and may destroy your python environment.
So, we recommend to use the virtual environment (e.g. pyenv, virtualenv, pipenv, etc.).
Please see Pytorch Geometirc document for more details.
python main.py
-
Configuration:
The settings for train and test are inconfig.yml
. -
Dataset:
Default training dataset is MovieLens-100k which will be automatically downloaded indata/
when runningmain.py
.
Note that the score of this repo donesn't reach to the original one.
Test RMSE | |
---|---|
Ours | 0.968 |
Original | 0.910 |