/Poetry

A RNN model to automatically generate Chinese poems.

Primary LanguagePython

Poetry

1. Introduction

A RNN model to automatically generate Chinese ancient poems with the input of start words. The idea is inspired by Weiyi Zheng's tangshi-rnn and Andrej Karpathy's Char-RNN.

2. Samples

1)             2)             3)             4)
白鹭窥鱼立，    一夜北风紧，    迟日江山丽，     去年今日此门中，
梅花晚少寒。    空山风雨寒。    无人入梦新。     白发春风未可怜。
青山犹出处，    山中有秋竹，    风霜独有路，     一笑一声天下去，
新月未无花。    风月更清香。    一月几人多。     自思春意又人心。
世事如收去，
相逢未自知。

3. Train and make poems

python main.py '一夜北风紧，' -p 100

Train a model If no available one exists or just load it and use the model to make poems. Generate a word2vec model and use the result to init the embedding layer weights of the rnn model. The model stacked 2 LSTM modules, each with 512 neurons. And 0.2 dropout rate while training. Please view the code to get more details.

4. Data

Only use a small part of Quan-Tang-Shi and Quan-Song-Shi.

5. Todo

A lot, such as adding rhythm.