This is a hackathon project for Hack UTD VI. It is no longer maintained.
At hackathons, you submit what is called a devpost at the end. it is basically a short blog post about what you did, why you did it, what technologies you used, etc. This project scraped a bunch of old devposts and used some machine learning to predict if a future project will be a winner, and also generate new fake hackathon projects.
Running is just running the site.py file in generate-text
directory with Python 3. That will start a local server on port 80 in
dev mode. You will need some Python dependencies which can be
installed with pip3, such as beautifulsoup4, flask, wtforms,
numpy, scikit-learn, and matplotlib. You can then navigate to
localhost for the randomly generated posts, and localhost/predict
to enter the title of a devpost to predict if that project will win.
Like all hackathon projects, everything is thrown together quickly and is a bit of a mess. We didn't always follow best practices with git and such, so there are definitely messes left around. Also, the sites are not generated as you refresh, rather pull from a pre-compiled list of data we created using gpt2.
N/A
Generates a devpost for a hackathon project, and predicts if a project is a winner based on its devpost.
First, we explored text generation by forcing a bot to read all of
Twilight, and then rewrite its own version. Meanwhile, Jake the data
fairy scraped all the most popular devposts from devpost.com, using
the Python and BeautifulSoup. Once the bot could write with the same
suspenseful elegance of Stephanie Meyers, we decided it was ready to
take on the devposts. Our text generating/sass department used
gpt-2-simple and ignored how little sense the results made.
Creating the model to classify a project was rough. After extensive data analysis, we came to the conclusion that there is a minimal correlation between any quantitative data that can be pulled from a devpost and the project's success. So that is a fat our bad. The prediction algorithm will classify a project with the same accuracy as that of a slightly more insightful coin. Someone kept moving things around in the git repo and Jake forgot his glasses. Also our bot got waaaaay too political.
- Successfully implemented k-nearest neighbors
- Generated adequately coherent text
- Had some good yucks
- There is NOT enough coconut water to go around
- The quality of a devpost is entirely subjective ;)
- Jeb Bush was in prison for 30 years
- We plan to continue by creating random manifesto generators and exploring writing our own novels by training our bot on individual author’s styles.