This project was made by 4 Le Wagon Data Science students as our final project. We have sourced and processed a wide range of stand-up scripts to analyse comedians and make our own comedy.

• Constructed a scraper to source and clean 3.6 million words of stand-up comedy from a variety of online sources using BeautifulSoup, Requests, and Pandas.
• Carried out a machine learning analysis using a Latent Dirichlet Allocation (LDA) on the processed dataset and used wordclouds to visualise results.
• Finetuned a bot using GPT2 (gpt-2-simple, credit: Max Woolf) on the subset of female comedian scripts & generted entirely original stand-up comedy material from only a few words of text input. Created an API for the bot using FastAPI, Docker and Google Cloud Run.
• Integrated and published the completed project on a public site using Heroku and Streamlit.

Heroku/Streamlit App

• Scraps From the Loft - Scripts and Year
• Subsaga - Scripts
• TMDB - The Movie Database - Artist Age and Gender

• 555 individual transcripts
• 268 comedians
• 19 million characters
• 3.6 million words

• Python
• Jupyter Notebook
• Requests + BeatuifulSoup
• Pandas
• NLTK
• Gensim
• GPT-2
• Docker
• Google Cloud AI Platform
• Google Cloud Run
• Heroku
• Streamlit

Full list of python packages can be found in the requirements.txt file.

• Reinis Melbardis @rmelbardis
• Yuqing Wang @yuqingwwang
• James Farrell @jfazz9
• Catriona Beamish @beamishc

• 0.1
  • Initial Release

• Le Wagon Data Science Course, London