Skip to content

Implementation of high-performance xDeepFM based on FuxiCTR.

Notifications You must be signed in to change notification settings

LiangcaiSu/xDeepFM_FuxiCTR

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

[FuxiCTR] CTR预估模型的高效、高性能实现(二)

本仓库是便于读者快速上手FuxiCTR而设立,具体实现步骤请参考原博文

FuxiCTR

仓库使用的是FuxiCTR v1.1版本,原仓库请参考:

# FuxiCTR dev version
https://github.com/xue-pai/FuxiCTR

FuxiCTR的版权归开发者所有,推荐使用最新版替换仓库中的fuxictr。

数据集

仓库提供的数据为采样数据,仅使用于熟悉和试用FuxiCTR框架。 处理过后的全量数据请参考BARS。 博文使用的数据为Criteo_x4_001。

运行脚本

依据配置文件训练一个模型:

nohup python -u main_xdeepfm.py --gpu 0 > logs/xdeepfm_criteo_x4.log & 

使用调参工具训练多个模型:

切换到目录tuner下(cd tuner)

nohup python -u run_param_tuner.py --config ../config/xdeepfm_criteo_x4/tuner_xdeepfm_criteo_x4.yaml --gpu 0 1 > ../logs/tuner_xdeepfm_criteo_x4.log & 

实验结果

以下为使用全量Criteo_x4数据的实验结果,仅供参考。

 20220104-055623,[command] python run_expid.py --version pytorch --config ../config/xdeepfm_criteo_x4/tuner_xdeepfm_criteo_x4 --expid xDeepFM_criteo_x4_001_1a332027 --gpu 0,[exp_id] xDeepFM_criteo_x4_001_1a332027,[dataset_id] criteo_x4,[train] N.A.,[val] logloss: 0.438579 - AUC: 0.813240,[test] logloss: 0.438173 - AUC: 0.813722
 20220104-095652,[command] python run_expid.py --version pytorch --config ../config/xdeepfm_criteo_x4/tuner_xdeepfm_criteo_x4 --expid xDeepFM_criteo_x4_005_ca239a4e --gpu 0,[exp_id] xDeepFM_criteo_x4_005_ca239a4e,[dataset_id] criteo_x4,[train] N.A.,[val] logloss: 0.439027 - AUC: 0.812783,[test] logloss: 0.438706 - AUC: 0.813151
 20220104-114355,[command] python run_expid.py --version pytorch --config ../config/xdeepfm_criteo_x4/tuner_xdeepfm_criteo_x4 --expid xDeepFM_criteo_x4_006_846e41f2 --gpu 1,[exp_id] xDeepFM_criteo_x4_006_846e41f2,[dataset_id] criteo_x4,[train] N.A.,[val] logloss: 0.438937 - AUC: 0.812833,[test] logloss: 0.438505 - AUC: 0.813312
 20220104-145342,[command] python run_expid.py --version pytorch --config ../config/xdeepfm_criteo_x4/tuner_xdeepfm_criteo_x4 --expid xDeepFM_criteo_x4_008_abbf6472 --gpu 0,[exp_id] xDeepFM_criteo_x4_008_abbf6472,[dataset_id] criteo_x4,[train] N.A.,[val] logloss: 0.438661 - AUC: 0.813146,[test] logloss: 0.438345 - AUC: 0.813516

About

Implementation of high-performance xDeepFM based on FuxiCTR.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages