An ML-based classical Chinese poem generator.

Update: This was migrated into Python3 and TensorFlow 1.8 in Jun 2018.

Update: This was migrated off TensorFlow 1.8 into PyTorch 2.5 in Jan 2025.

• Vocab: A set of characters and their pre-trained embeddings.

• Generator: A poem generator based on LSTM encoder-decoder with Bahdanau attention.

• Discriminator: An LSTM-based classifier that distinguishes real poems from fake ones.

• SeqGAN Trainer: A training program that implements the idea of the SeqGAN model. This is still in the experimental stage.

python3 -m venv venv
source venv/bin/activate
pip3 install -r requirements.txt

To train the word embedding model:

python3 vocab.py

To train the generator model:

python3 generator.py

To train the discriminator model:

python3 discriminator.py

To train the SeqGAN model:

python3 train_seq_gan.py

Start the program with:

python3 demo.py

Then enter a sequence of Chinese characters after the prompt as the first characters of the poem sentences to be generated.

Generate a poem given the first character of each sentence (藏头诗):

>>> 风和日丽鸟语花香
风雨萧萧秋月明
和风吹雨湿衣裳
日暮东风吹柳絮
丽晴春色满天涯
鸟啼花落春风里
语语春风满袖风
花落花开不知处
香风吹落柳丝丝

>>> 精诚之至金石为开
精神禀气在天涯
诚是中原万事非
之子何人知此意
至今古道无人知
金丸玉骨不可识
石泉泉石无人知
为君一醉一杯酒
开尽桃花一万株

Search similar characters to a given character:

风: 飙飚吹飕飒
花: 蕊红葩杏桃
雪: 霜霰雨腊絮
月: 蟾影皎夜魄
鸟: 禽噪鸦鹂鸥
鱼: 鳞鲤鲈脍虾
虫: 蛩唧蟀啾蟋