Hierarchical-Attention-Network

HAN model. Three versions.

• version 1: Tensorflow, dynamic GRU
• version 2: Pytorch, pack_padded_sequence
• version 3: Pytorch, mask matrix

Paper

Hierarchical Attention Network (readed in 2017/10 by szx)

Task Instruction

• Text Classification

Code Instruction

• data: API for loading text data

• models: HAN model

  • hierarchical_tf.py : version 1
  • hierarchical_pack.py : version 2
  • hierarchical_mask.py : version 3

• preprocessor: Data preprocessing

• utils: tools for training

• config.py: hyperparameters for models

Run code

CUDA_VISIBLE_DEVICES=X python train.py --model-id 4 --is-save (y or n)

Write in the end

• The model was implemented in BDCI2017.

• We implement eight models, such as FastText, TextCNN, TextRNN, RCNN, HAN, CNNInception, CNNWithDoc2Vec, RCNNWithDoc2Vec.

• I just make contribution to HAN model.

• Thanks for my teammates, lhliu, dyhu.