Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/xinyu-intel/ncf_mxnet
Neural Collaborative Filtering with MXNet
https://github.com/xinyu-intel/ncf_mxnet
collaborative-filtering deep-learning embedding-models mxnet ncf recommendation-system recommender-system
Last synced: about 1 month ago
JSON representation
Neural Collaborative Filtering with MXNet
- Host: GitHub
- URL: https://github.com/xinyu-intel/ncf_mxnet
- Owner: xinyu-intel
- Created: 2019-07-05T13:46:09.000Z (over 5 years ago)
- Default Branch: master
- Last Pushed: 2019-12-23T07:55:58.000Z (almost 5 years ago)
- Last Synced: 2024-11-08T18:46:33.104Z (about 1 month ago)
- Topics: collaborative-filtering, deep-learning, embedding-models, mxnet, ncf, recommendation-system, recommender-system
- Language: Python
- Homepage:
- Size: 27.4 MB
- Stars: 4
- Watchers: 1
- Forks: 2
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- Awesome-MXNet - ncf
README
# Neural Collaborative Filtering
[](https://travis-ci.com/xinyu-intel/ncf_mxnet)
This is MXNet implementation for the paper:
Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu and Tat-Seng Chua (2017). [Neural Collaborative Filtering.](http://dl.acm.org/citation.cfm?id=3052569) In Proceedings of WWW '17, Perth, Australia, April 03-07, 2017.
Three collaborative filtering models: Generalized Matrix Factorization (GMF), Multi-Layer Perceptron (MLP), and Neural Matrix Factorization (NeuMF). To target the models for implicit feedback and ranking task, we optimize them using log loss with negative sampling.
Author: Dr. Xiangnan He (http://www.comp.nus.edu.sg/~xiangnan/)
Code Reference: https://github.com/hexiangnan/neural_collaborative_filtering
## Environment Settings
We use MXnet with MKL-DNN as the backend.
- MXNet version: MXNet Master(TBD)
## Install
```
pip install -r requirements.txt
```
## Dataset
We provide the processed datasets on [Google Drive](https://drive.google.com/drive/folders/1qACR_Zhc2O2W0RrazzcepM2vJeh0MMdO?usp=sharing): MovieLens 20 Million (ml-20m), you can download directly or
run the script to prepare the datasets:
```
python convert.py ./data/
```
train-ratings.csv
- Train file (positive instances).
- Each Line is a training instance: userID\t itemID\t
test-ratings.csv
- Test file (positive instances).
- Each Line is a testing instance: userID\t itemID\t
test-negative.csv
- Test file (negative instances).
- Each line corresponds to the line of test.rating, containing 999 negative samples.
- Each line is in the format: userID,\t negativeItemID1\t negativeItemID2 ...
## Pre-trained models
We provide the pretrained ml-20m model on [Google Drive](https://drive.google.com/drive/folders/1qACR_Zhc2O2W0RrazzcepM2vJeh0MMdO?usp=sharing), you can download directly for evaluation or calibration.
|dtype|HR@10|NDCG@10|
|:---:|:--:|:--:|
|float32|0.6393|0.3849|
|float32 opt|0.6393|0.3849|
|int8|0.6395|0.3852|
|int8 opt|0.6396|0.3852|
## Training
```
# train ncf model with ml-20m dataset
python train.py # --gpu=0
```
## Model Optimizer
```
# optimize nc model
python model_optimizer.py
```
## Calibration
```
# neumf calibration on ml-20m dataset
python ncf.py --prefix=./model/ml-20m/neumf --calibration
# optimized neumf calibration on ml-20m dataset
python ncf.py --prefix=./model/ml-20m/neumf-opt --calibration
```
## Evaluation
```
# neumf float32 inference on ml-20m dataset
python ncf.py --batch-size=1000 --prefix=./model/ml-20m/neumf
# optimized neumf float32 inference on ml-20m dataset
python ncf.py --batch-size=1000 --prefix=./model/ml-20m/neumf-opt
# neumf int8 inference on ml-20m dataset
python ncf.py --batch-size=1000 --prefix=./model/ml-20m/neumf-quantized
# optimized neumf int8 inference on ml-20m dataset
python ncf.py --batch-size=1000 --prefix=./model/ml-20m/neumf-opt-quantized
```
## Benchmark
```
usage: bash ./benchmark.sh [[[-p prefix ] [-e epoch] [-d dataset] [-b batch_size] [-i instance] [-c cores/instance]] | [-h]]
# neumf float32 benchmark on ml-20m dataset
sh benchmark.sh -p model/ml-20m/neumf
# optimized neumf float32 benchmark on ml-20m dataset
sh benchmark.sh -p model/ml-20m/neumf-opt
# neumf int8 benchmark on ml-20m dataset
sh benchmark.sh -p model/ml-20m/neumf-quantized
# optimized neumf int8 benchmark on ml-20m dataset
sh benchmark.sh -p model/ml-20m/neumf-opt-quantized
```