Twitter Transfer Learning

Create twitter bots in-browser using Tensorflow.js and transfer learning with a pre-trained Keras RNN model from brangerbriz/char-rnn-text-generation.

Download & Install

# clone the project, including submodules
git clone --recursive https://github.com/brangerbriz/twitter-transfer-learning
cd twitter-transfer-learning

# install dependencies
npm install

# if you have an NVIDIA graphics card and CUDA installed, download tfjs-node-gpu
# If you do not, or don't know what this means, skip this step...
npm install @tensorflow/tfjs-node-gpu
# replace all instances of @tensorflow/tfjs-node with @tensorflow/tfjs-node-gpu in src/
find src/ -type f | xargs sed -i "s#require(['\"\`]@tensorflow/tfjs-node['\"\`])#require('@tensorflow/tfjs-node-gpu')#g"

Install Tweet-Server

This project relies on brangerbriz/tweet-server. To run it, you must have an instance of tweet-server running on localhost port 3000, which will facilitate the downloading of twitter data for this application.

# clone the repo
git clone https://github.com/brangerbriz/tweet-server
cd tweet-server

# install the dependencies
npm install

# rename the example env file
cp .env-example .env

# ... edit .env using your own twitter API credentials

# start the server
npm start

Run

cd twitter-transfer-learning
npm start

This will launch an electron application that you can use to:

1. Download twitter data for a user
2. Train a model using that user's twitter data, leveraging a pre-trained keras "base-model" that's been trained on > 9,000,000 tweets.
3. Generate new tweets with your trained model.

WARNING: Model training and tweet generation can a while depending on your computer's resources.

License

Copyright (C) 2018 Branger_Briz

Licensed under the GPL v3 or any later version.