Counting 3,222 Big Data & Machine Learning Frameworks, Toolsets, and Examples...
Suggestion? Feedback? Tweet @stkim1

Last Commit
Sep. 22, 2018
Sep. 17, 2017

CNN-DCNN text autoencoder

Implementations of the models in the paper "Deconvolutional Paragraph Representation Learning" by Yizhe Zhang, Dinghan Shen, Guoyin Wang, Zhe Gan, Ricardo Henao and Lawrence Carin, NIPS 2017


  • CUDA, cudnn
  • Tensorflow (version >1.0). We used tensorflow 1.2. Run: pip install -r requirements.txt to install requirements


  • Run: python for reconstruction task
  • Run: python for character-level correction task
  • Run: python for semi-supervised task
  • Options: options can be made by changing option class in the code.
  • opt.n_hidden: number of hidden units.
  • opt.layer: number of CNN/DCNN layer [2,3,4].
  • learning rate.
  • opt.batch_size: number of batchsize.
  • Training roughly takes 6-7 hours (around 10-20 epochs) (for recontruction task) to converge on a K80 GPU machine.
  • See output.txt for a sample of screen output for reconstruction task.



Please cite our paper if it helps with your research

  title={Deconvolutional Paragraph Representation Learning},
  author={Zhang, Yizhe and Shen, Dinghan and Wang, Guoyin and Gan, Zhe and Henao, Ricardo and Carin, Lawrence},

For any question or suggestions, feel free to contact [email protected]