Hierachical Attention Networks

An efficient pytorch implementation of Hierarchical Attention Networks model proposed in the paper https://www.cs.cmu.edu/~hovy/papers/16HLT-hierarchical-attention-networks.pdf

Repository includes all supporting files along with the model as described below:

• data.py - Reads data from a pandas dataframe containing fields 'text' and 'label'. Expects the data to be split into train, test and val each into separate csv files.
• util.py - Contains batcher methods, embedding methods etc.
• train.py - Contains train and eval methods.
• main.py - Contains the main training loop. Specify data paths, embedding files and other settings here.
• model.py - Contains the model definition.

Dependencies

• gensim
• sklearn
• pytorch