songci-rnn

Generating Songci (Poetry of the Song Dynasty) by machine. Based on QuanSongci and RNN.
Live Demo
If the sample result is shit, you can refresh the page to make it surprise you.

---

How to use

1. Get torch-rnn. torch-rnn is based on Torch7, you should set up Torch environment first.
2. Copy the songci.txt to torch-rnn/data/
3. Enter torch-rnn directory
4. You can train your own model or use my provided model. Run

   python scripts/preprocess.py --input_txt data/songci.txt --output_h5 songci.h5 --output_json songci.json
   
   # train the model
   th train.lua -input_h5 songci.h5 -input_json songci.json -gpu -1  # without gpu
   
   # sample text
   th sample.lua -checkpoint cv/checkpoint_yourcheckpoint.t7 -length 200 -gpu -1
   
   # or you can sample text from my trained model
    th sample.lua -checkpoint songci_epoch50_loss5.106.t7 -length 200 -gpu -1
   

吟·宋词

通过全宋词训练RNN，随机生成宋词
Live Demo
如果生成结果很烂，可尝试多刷新几次，总会惊艳到你。有些反正我是不相信这竟然是机器生成的（= =）

---

使用

1. 下载 torch-rnn. torch-rnn 基于 Torch7, 首先先配置Torch环境
2. 拷贝 songci.txt 到 torch-rnn/data/
3. 进入 torch-rnn 目录
4. 你可以自己训练模型也可以用我提供的模型

   python scripts/preprocess.py --input_txt data/songci.txt --output_h5 songci.h5 --output_json songci.json
   
   # 训练
   th train.lua -input_h5 songci.h5 -input_json songci.json -gpu -1  # cpu模式
   
   # 生成文本
   th sample.lua -checkpoint cv/checkpoint_yourcheckpoint.t7 -length 200 -gpu -1
   
   # 用我提供的模型生成文本
    th sample.lua -checkpoint songci_epoch50_loss5.106.t7 -length 200 -gpu -1
   

If you have any problem, please contact hevlhayt@foxmail.com (ﾉﾟ▽ﾟ)ﾉ