GAN-based Matrix Factorization for Recommender Systems

Overview

GAN-based Matrix Factorization for Recommender Systems

GANMF architecture

This repository contains the datasets' splits, the source code of the experiments and their results for the paper "GAN-based Matrix Factorization for Recommender Systems" (arXiv: https://arxiv.org/abs/2201.08042) accepted at the 37th ACM/SIGAPP Symposium on Applied Computing (SAC '22).

How to use this repo

This repo is based on a version of Recsys_Course_AT_PoliMi. In order to run the code and experiments you need first to setup a Python environment. Any environment manager will work, but we suggest conda since it is easier to recreate our environment if using a GPU. conda can help with the installation of CUDA and CUDA toolkit necessary to utilize available GPU(s). We highly recommend running this repo with a GPU since GAN-based recommenders require long training times.

Conda

Run the following command to create a new environment with Python 3.6.8 and install all requirements in file conda_requirements.txt:

conda create -n <name-env> python==3.6.8 --file conda_requirements.txt

The file conda_requirements.txt also contains the packages cudatoolkit==9.0 and cudnn==7.1.2 which are installed completely separate from other versions you might already have installed and are managed by conda.

Install the following packages using pip inside the newly created environment since they are not found in the main channel of conda and conda-forge channel holds old versions of them:

pip install scikit-optimize==0.7.2 telegram-send==0.25

Activate the newly created environment:

conda activate <name-env>

Virtualenv & Pip

First download and install Python 3.6.8 from python.org. Then install virtualenv:

python -m pip install --user virtualenv

Now create a new environment with virtualenv (by default it will use the Python version it was installed with):

virtualenv <name-env> <path-to-new-env>

Activate the new environment with:

source <path-to-new-env>/bin/activate

Install the required packages through the file pip_requirements.txt:

pip install -r pip_requirements.txt

Note that if you intend to use a GPU and install required packages using virtualenv and pip then you need to install separately cudatoolkit==9.0 and cudnn==7.1.2 following instructions for your GPU on nvidia.com.

Before running any experiment or algorithm you need to compile the Cython code part of some of the recommenders. You can compile them all with the following command:

python run_compile_all_cython.py

N.B You need to have the following packages installed before compiling: gcc and python3-dev.

N.B Since the experiments can take a long time, the code notifies you on your Telegram account when the experiments start/end. Either configure telegram-send as indicated on https://pypi.org/project/telegram-send/#installation or delete the lines containing telegram-send inside RecSysExp.py.


Running experiments

All results presented in the paper are already provided in this repository. In case you want to re-run the experiments, below you can find the steps for each one of them.

Comparison with baselines1

In order to run all the comparisons with the baselines use the file RecSysExp.py. First compute for each dataset the 5 mutually exclusive sets:

  • Training set: once best hyperparameters of the recommender are found, it will be finally trained with this set.

    • Training set small: the recommender is first trained on this small training set with the aim of finding the best hyperparameters.
    • Early stopping set: validation set used to incorporate early stopping in the hyperparameters tuning.
    • Validation set: the recommender with the current hyperparameter values is tested against this set.
  • Test set: once the best hyperparameters are found, the recommender is finally tested with this set. The results presented are the ones on this set.

Compute the splits for each dataset with the following command:

python RecSysExp.py --build-dataset <dataset-name>

To run the tuning of a recommender use the following command:

python RecSysExp.py <dataset-name> <recommender-name> [--user | --item] [<similarity-type>] 
  • dataset-name is a value among: 1M, hetrec2011, LastFM.
  • recommender-name is a value among: TopPop, PureSVD, ALS, SLIMBPR, ItemKNN, P3Alpha, CAAE, CFGAN, GANMF.
  • --user or --item is a flag used only for GAN-based recommenders. It denotes the user/item-based training procedure for the selected recommender.
  • similarity-type is a value among: cosine, jaccard, tversky, dice, euclidean, asymmetric. It is used only for ItemKNN recommender.

All results, best hyperparameters and dataset splits are saved in the experiments directory.


Testing on test set with best hyperparameters

In order to test each tuned recommender on the test set (which is created when tuning the hyperparameters) run the following command:

python RunBestParameters.py <dataset-name> <recommender-name> [--user | --item] [<similarity-type>] [--force] [--bp <best-params-dir>]
  • dataset-name is a value among: 1M, hetrec2011, LastFM.
  • recommender-name is a value among: TopPop, PureSVD, ALS, SLIMBPR, ItemKNN, P3Alpha, CAAE, CFGAN, GANMF.
  • --user or --item is a flag used only for GAN-based recommenders. It denotes the user/item based training procedure for the selected recommender.
  • similarity-type is a value among: cosine, jaccard, tversky, dice, euclidean, asymmetric. It is used only for ItemKNN recommender.
  • --force is a flag that forces the computation of the results on test set. By default, if the result for the tuple (dataset, recommender) exists in test_result directory, the computation is not performed.
  • --bp sets the directory where the best parameters (best_params.pkl) are located for this combination of (dataset, recommender), by default in experiments directory.

The results are saved in the test_results directory.


Ablation study

To run the ablation study, use the script AblationStudy.py as follows:

python AblationStudy.py <dataset-name> [binGANMF | feature-matching [--user | --item]]
  • dataset-name is a value among: 1M, hetrec2011, LastFM.
  • binGANMF runs the first ablation study, the GANMF model with binary classifier discrimnator. This tunes the recommender with RecSysExp.py and then evaluates it with RunBestParameters.py on the test set.
  • --user or --item is a flag that sets the training procedure for binGANMF recommender.
  • feature-matching runs the second ablation study, the effect of the feature matching loss and the user-user similarity heatmaps. The results are saved in the feature_matching directory.

MF model of GANMF

To run the qualitative study on the MF learned by GANMF, use the script MFLearned.py as follows:

python MFLearned.py

It executes both experiments and the results are saved in the latent_factors directory.

Footnotes

  1. For the baselines Top Popular, PureSVD, ALS, SLIMBPR, ItemKNN, P3Alpha and model evaluation we have used implementations from Recsys_Course_AT_PoliMi.

Owner
Ervin Dervishaj
Interested in Recommender Systems and Machine/Deep Learning research
Ervin Dervishaj
Hardware accelerated, batchable and differentiable optimizers in JAX.

JAXopt Installation | Examples | References Hardware accelerated (GPU/TPU), batchable and differentiable optimizers in JAX. Installation JAXopt can be

Google 621 Jan 08, 2023
PixelPick This is an official implementation of the paper "All you need are a few pixels: semantic segmentation with PixelPick."

PixelPick This is an official implementation of the paper "All you need are a few pixels: semantic segmentation with PixelPick." [Project page] [Paper

Gyungin Shin 59 Sep 25, 2022
An example to implement a new backbone with OpenMMLab framework.

Backbone example on OpenMMLab framework English | 简体中文 Introduction This is an template repo about how to use OpenMMLab framework to develop a new bac

Ma Zerun 22 Dec 29, 2022
A sequence of Jupyter notebooks featuring the 12 Steps to Navier-Stokes

CFD Python Please cite as: Barba, Lorena A., and Forsyth, Gilbert F. (2018). CFD Python: the 12 steps to Navier-Stokes equations. Journal of Open Sour

Barba group 2.6k Dec 30, 2022
Awesome Graph Classification - A collection of important graph embedding, classification and representation learning papers with implementations.

A collection of graph classification methods, covering embedding, deep learning, graph kernel and factorization papers

Benedek Rozemberczki 4.5k Jan 01, 2023
Generic image compressor for machine learning. Pytorch code for our paper "Lossy compression for lossless prediction".

Lossy Compression for Lossless Prediction Using: Training: This repostiory contains our implementation of the paper: Lossy Compression for Lossless Pr

Yann Dubois 84 Jan 02, 2023
CryptoFrog - My First Strategy for freqtrade

cryptofrog-strategies CryptoFrog - My First Strategy for freqtrade NB: (2021-04-20) You'll need the latest freqtrade develop branch otherwise you migh

Robert Davey 137 Jan 01, 2023
[CVPR 2021] Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers

[CVPR 2021] Rethinking Semantic Segmentation from a Sequence-to-Sequence Perspective with Transformers

Fudan Zhang Vision Group 897 Jan 05, 2023
Pytorch implementations of popular off-policy multi-agent reinforcement learning algorithms, including QMix, VDN, MADDPG, and MATD3.

Off-Policy Multi-Agent Reinforcement Learning (MARL) Algorithms This repository contains implementations of various off-policy multi-agent reinforceme

183 Dec 28, 2022
PyBrain - Another Python Machine Learning Library.

PyBrain -- the Python Machine Learning Library =============================================== INSTALLATION ------------ Quick answer: make sure you

2.8k Dec 31, 2022
OpenPose: Real-time multi-person keypoint detection library for body, face, hands, and foot estimation

Build Type Linux MacOS Windows Build Status OpenPose has represented the first real-time multi-person system to jointly detect human body, hand, facia

25.7k Jan 09, 2023
MetaAvatar: Learning Animatable Clothed Human Models from Few Depth Images

MetaAvatar: Learning Animatable Clothed Human Models from Few Depth Images This repository contains the implementation of our paper MetaAvatar: Learni

sfwang 96 Dec 13, 2022
A set of tests for evaluating large-scale algorithms for Wasserstein-2 transport maps computation.

Continuous Wasserstein-2 Benchmark This is the official Python implementation of the NeurIPS 2021 paper Do Neural Optimal Transport Solvers Work? A Co

Alexander 22 Dec 12, 2022
NeoPlay is the project dedicated to ESport events.

NeoPlay is the project dedicated to ESport events. On this platform users can participate in tournaments with prize pools as well as create their own tournaments.

3 Dec 18, 2021
Hand tracking demo for DIY Smart Glasses with a remote computer doing the work

CameraStream This is a demonstration that streams the image from smartglasses to a pc, does the hand recognition on the remote pc and streams the proc

Teemu Laurila 20 Oct 13, 2022
Github for the conference paper GLOD-Gaussian Likelihood OOD detector

FOOD - Fast OOD Detector Pytorch implamentation of the confernce peper FOOD arxiv link. Abstract Deep neural networks (DNNs) perform well at classifyi

17 Jun 19, 2022
Randomized Correspondence Algorithm for Structural Image Editing

===================================== README: Inpainting based PatchMatch ===================================== @Author: Younesse ANDAM @Conta

Younesse 116 Dec 24, 2022
Awesome Remote Sensing Toolkit based on PaddlePaddle.

基于飞桨框架开发的高性能遥感图像处理开发套件,端到端地完成从训练到部署的全流程遥感深度学习应用。 最新动态 PaddleRS 即将发布alpha版本!欢迎大家试用 简介 PaddleRS是遥感科研院所、相关高校共同基于飞桨开发的遥感处理平台,支持遥感图像分类,目标检测,图像分割,以及变化检测等常用遥

146 Dec 11, 2022
Drone-based Joint Density Map Estimation, Localization and Tracking with Space-Time Multi-Scale Attention Network

DroneCrowd Paper Detection, Tracking, and Counting Meets Drones in Crowds: A Benchmark. Introduction This paper proposes a space-time multi-scale atte

VisDrone 98 Nov 16, 2022
Wandb-predictions - WANDB Predictions With Python

WANDB API CI/CD Below we capture the CI/CD scenarios that we would expect with o

Anish Shah 6 Oct 07, 2022