Skip to content
/ DeepCTR Public
forked from shenweichen/DeepCTR

Easy-to-use,Modular and Extendible package of deep-learning based CTR models.DeepFM,DeepInterestNetwork(DIN),DeepCrossNetwork(DCN),AttentionalFactorizationMachine(AFM),Neural Factorization Machine(NFM),AutoInt

License

Notifications You must be signed in to change notification settings

jkfdre/DeepCTR

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

3 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

机器学习模型库说明文档

依赖环境

  • python3
  • tensorflow
  • numpy
  • sklearn

设计说明

base基类仿照keras模型实现以下公有方法,包括

  • save_model 保存模型
  • load_mdel 加载模型
  • train_on_batch 小批量训练
  • fit 全量训练
  • test_on_batch 小批量评估
  • evaluate 模型评估
  • predict_on_batch 小批量预测
  • predict 全量预测

同时设计了若干抽象方法

  • _get_data_loss
  • _get_input_target
  • _get_output_target
  • _get_data_loss
  • _get_optimizer
  • _build_graph

要求子类在__init__方法的最后调用self._build_graph()构建计算图。

DeepFM

DeepFM: A Factorization-Machine based Neural Network for CTR Prediction arxiv

DeepCrossNetwork

Deep & Cross Network for Ad Click Predictions arxiv

实现和论文的区别

  • embedding_size
    论文里提出每个field根据cardinality的不同来设置。这里用的是所有field具有相同embedding_size的实现,所以feature的编码是global的。
  • Batch normalization
    论文在Deep Network部分采用了BN
  • gradient clip
    论文采用了梯度截断,范数设置为100
  • 其他 论文采用512batch size,并提出使用早停来防止过拟合,L2和dropout正则效果一般。

About

Easy-to-use,Modular and Extendible package of deep-learning based CTR models.DeepFM,DeepInterestNetwork(DIN),DeepCrossNetwork(DCN),AttentionalFactorizationMachine(AFM),Neural Factorization Machine(NFM),AutoInt

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 100.0%