A set of scripts for automating the torch-rnn preprocess, training and sampling process. This is based off torch-rnn (https://github.com/crisbal/docker-torch-rnn) and the Docker images thereof (https://github.com/crisbal/docker-torch-rnn)
- A machine running Linux (OSX might also work, but is untested)
- Docker
- Some sample data, the more the better
- Put your source data into the
source_datafolder - Run
./run_all.sh. - For further samples, simply run
./sample.sh
Please note that training can and will take a long time and consume a lot of processing power. Checkpoints are saved every 200 iterations, so it's reccommended you Ctrl+C to stop the training early to use the most recent checkpoint.
- Put your source data into the
source_datafolder - (Optional) Run flatten_source_data.sh to flatten the directory structure in source_data
- Preprocess using
./preprocess.sh - Train using
./train.sh - After 200 iterations or more, stop the training with Ctrl+C
- Generate samples using
./sample.sh