What it does
Generates a devpost for a hackathon project, and predicts if a project is a winner based on its devpost.
How I built it
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.
Challenges I ran into
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.
Accomplishments that I'm proud of
- Successfully implemented k-nearest neighbors
- Generated adequately coherent text
- Had some good yucks
What I learned
- There is NOT enough coconut water to go around
- The quality of a devpost is entierly subjective ;)
- Jeb Bush was in prison for 30 years
What's next for This Hackathon Project is Not Real
- We plan to continue by creating random manifesto generators and exploring writing our own novels by training our bot on individual author’s styles.