Character-based recurrant neural networks from colloquial English with a touch of Shakespeare
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.


This code has been deprecated in favor of sequelspeare

Character-based recurrent neural networks wrapped into an IRC chatbot. Initially designed to be colloquial English with a touch of Shakespeare, the network can't be open-sourced for privacy reasons: training data came from IRC users who didn't consent to open-sourcing.

tl;dr environment setup

Swiggityspeare now comes with an easy setup script to get started! You'll still need to install CUDA manually, but here's the process:

  • install the CUDA Toolkit
  • git clone --recursive
  • cd swiggityspeare
  • ./
  • java -jar swiggityspeare.jar -c "#swag #cwru" -n babbyspeare will ask to install Torch, init+update the git submodule for Andrej Karpathy's char-rnn, move the neural network into place, and install the required packages with luarocks.


Swiggity's Java dependencies are all taken care of at compile time with the .jar files in the dependencies directory.

Other than CUDA, all of swiggity's system dependencies are best installed using, as explained above. Otherwise, manually install Torch, git submodule init and git submodule update, and use luarocks (provided with Toch) to install cunn cutorch nngraph optim.

Test the environment with a quick th train.lua, whose default settings should begin training from a Shakespeare dataset included in the char-rnn repository.

It's build-your-own neural network. A sample network is in dependencies/irc_network.t7: it's a quick Shakespeare-trained network. This needs to be moved into the newly cloned char-rnn directory to work. manages the initial copying for you.

At this point usage is pretty simple. The project is based on IntelliJ, so just build->build artifacts->swiggityspeare.jar, or use the prebuilt swiggityspeare.jar provided.


Execution is a simple java -jar swiggityspeare.jar -n bot_nick. However, the jar expects the dependencies directory to be alongside it (or the configuration to be set), so that it knows where the char-rnn code is, and can execute it. The prepackaged .t7 neural network needs to be moved into the cloned char-rnn directory.

When in doubt, read one of the .prop files for options. A new one is generated for each bot nick.


Git-workflow-style fork-commit-pullreqest. You kids are all better at git than I am, so I'm learning from you here.