Introduction

This is a PyTorch implementation of A New Method of Region Embedding for Text Classification (ICLR 2018). MR dataset is used by default.

NOTE: in this implementation, only Context-Word Region Embedding is available for now. I do not have a plan to implement Word-Context Region Embedding in the intermediate future. If anyone has a chance to implement the former, please let me know. Thanks in advance.

I also added a self-attention layer on top of the region embedding layer to get better results.

Requirements

• python 3
• PyTorch >= 0.3
• sklearn
• numpy

Usage

python main.py [arguments]

Arguments

-h, -help       help information
-embed_dim      dimension of word embeddings
-epochs         number of epoches