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.

Prerequisites

• Python 3.5
• tensorflow-gpu==1.3.0
• matplotlib==2.0.2
• numpy==1.13.1
• scikit-learn==0.19.0

Preparation

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,
                ...}

Usage

Simple VAE

Train

1. modify config.py
2. run

  python train_vae.py

Get sample sentences

1. modify sampling.py
2. run

  python sampling.py

Semisupervised Classification

1. modify config.py
2. run

  python train_cvae.py

License

MIT

Author

Ryo Kamoi