/AutoAttention

Tensorflow implementaion for AutoAttention

Primary LanguagePythonApache License 2.0Apache-2.0

AutoAttention

Requirements

Please use pip install -r requirements.txt to setup the operating environment in python3.5.
Note that we use DeepCTR package and refer to the implementation of DSIN.

Prepare data

  1. Download Alimama Data: Ad Display/Click Data on Taobao.com
  2. Extract the files into the data/raw_data directory
  3. Follow the code of the data preprocessing in DSIN to preprocess data

Training and Evaluation

Run python train_[model].py [model_type]

  • Sum Pooling: python train_base.py Base
  • MAF-S: python train_base.py Base_All_Fields_Add
  • MAF-C: python train_base.py Base_All_Fields_Concat
  • DIN: python train_din.py DIN
  • DIN+: python train_din.py All_Fields
  • DIEN: python train_dien.py
  • DSIN: python train_dsin.py
  • DotProduct: python train_autoattention.py DotProduct
  • AutoAttention: python train_autoattention.py AutoAttention_Prun 0.6