jiangnanhugo/lmkit

language models toolkits with hierarchical softmax setting

lmkit

1. original: Recurrent Language Model (RNNLM) with LSTM/GRU cells
2. Sampling: noise contrastive estimation/negative sampling/blackout for RNNLM
3. cHSM: class-based hierarchical softmax for RNNLM
4. masked-cHSM: unequal partitioned vocabulary case for class decomposition.
5. p-tHSM: paralleled tree-based hierarchical softmax for RNNLM
6. tHSM: traditional tree-based hierarchical softmax with Huffman coding for RNNLM

Accepted paper

@inproceedings{DBLP:conf/ijcai/JiangRGSX17,
  author    = {Nan Jiang and
               Wenge Rong and
               Min Gao and
               Yikang Shen and
               Zhang Xiong},
  title     = {Exploration of Tree-based Hierarchical Softmax for Recurrent Language
               Models},
  booktitle = {Proceedings of the Twenty-Sixth International Joint Conference on
               Artificial Intelligence, {IJCAI} 2017, Melbourne, Australia, August
               19-25, 2017},
  pages     = {1951--1957},
  year      = {2017},
  crossref  = {DBLP:conf/ijcai/2017},
  url       = {https://doi.org/10.24963/ijcai.2017/271},
  doi       = {10.24963/ijcai.2017/271},
  timestamp = {Tue, 15 Aug 2017 14:48:05 +0200},
  biburl    = {https://dblp.org/rec/bib/conf/ijcai/JiangRGSX17},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}