onuratakan/deeptweets

Generate tweets using GPT-2 in the style of famous politicians.

The project is tested on python 3.6.9, and it may work on later versions as well.

Create a virtual environment using the command

python -m venv deep_tweets_workspace

Clone this project inside the environment.

Dependencies

• Tensorflow
• Numpy
• Pandas
• PyTorch
• Rake
• Transformers
• Wandb (Weights and Biases)

Dependencies can be installed using pip:

pip install -r requirements.txt

Note, you'll need free a Weights and Biases account to run the GPT-2 models. The LSTM models can be run without it. Getting started: https://wandb.ai/

LSTM models

To run the inference on the LSTM model, navigate to the ./models/LSTM folder, and run the command

python inference.py [twitter_handle] [prompt]

To train the LSTM model from scratch, run the following command from the same directory

python training.py [twitter_handle]

twitter_handle can be either realDonaldTrump or JoeBiden, and prompt is the starting prompt of the sentence, e.g., I like

Note: Training the LSTM models take a significant amount of time. If you just want to obtain the results, run the inference.

GPT-2 models

Navigate to the directory GPT2, and run the file deeptweets_prompt.ipynb. for the DeepTweets-Prompt model and the file deeptweets_context.ipynb. for the DeepTweets-Context model. You can change the following line in each file to replace the handle:

handle = [twitter_handle]