twain-speak
Trains a character-based language model on novels using recurrent neural networks, to generate text in the style of Mark Twain.
Setup
Pull the docker repo
docker pull mbartoli/char-rnn
and then open a shell within the container
docker run -i -t CONTAINER_ID /bin/bash
Then, in the new shell, run
luarocks install nngraph
luarocks install optim
Next, clone this repo, and run setup.sh.
Training
To train, run
sh train.sh
Sampling
To generate text, edit sampling.sh and replace ENTER_CHECKPOINT_HERE with your model checkpoint found in /home/char-rnn/cv. Then run
sh sampling.sh
Additional Info
See
https://github.com/karpathy/char-rnn
http://karpathy.github.io/2015/05/21/rnn-effectiveness/
Docker
https://github.com/mbartoli/docker-char-rnn
https://registry.hub.docker.com/u/mbartoli/char-rnn/