TensorFlow implementation of "Improved Variational Autoencoders for Text Modeling using Dilated Convolutions"
paper:https://arxiv.org/abs/1702.08139v2
This is NOT an original implementation. There may be some minor differences from the original structure.
- Python 3.5
- tensorflow-gpu==1.3.0
- matplotlib==2.0.2
- numpy==1.13.1
- scikit-learn==0.19.0
Dataset is not contained. Please prepare your own dataset.
- Sentence
Pickle file of Numpy array of word ids (shape=[batch_size, sentence_length]).
- Label
Pickle file of Numpy array of a label of a class (sentiment, category, etc.) (shape=[batch_size]).
- Dictionary
Pickle file of Python dictionary. It should contain "<EOS>", "<PAD>", "<GO>" as meta words.
dictionary = {word1: id1,
word2: id2,
...}
- modify config.py
- run
python train_vae.py
- modify sampling.py
- run
python sampling.py
- modify config.py
- run
python train_cvae.py
MIT
Ryo Kamoi