Inspiration
Trump's statements include some of the most outrageous things said recently, so we wanted to see whether someone could distinguish between a fake statement and something Trump would say.
What it does
We generated statements using markov chains (https://en.wikipedia.org/wiki/Markov_chain) that are based off of the things trump actually says. To show how difficult it is to distinguish between the machine generated text and the real stuff he says, we made a web app to test whether someone could determine which was the machine generated text (Drumpf) and which was the real Trump speech.
How we built it
python+regex for parsing Trump's statementsurrent tools you use to find & apply to jobs?
html/css/js frontend azure and aws for backend/hosting
Challenges we ran into
Machine learning is hard. We tried to use recurrent neural networks at first for speech but realized we didn't have a large enough data set, so we switched to markov chains which don't need as much training data, but also have less variance in what is generated.
We actually spent a bit of time getting https://github.com/jcjohnson/torch-rnn up and running, but then figured out it was actually pretty bad as we had a pretty small data set (<100kB at that point)
Eventually got to 200kB and we were half-satisfied with the results, so we took the good ones and put it up on our web app.
Accomplishments that we're proud of
First hackathon we've done where our front end looks good in addition to having a decent backend. regex was interesting.
What we learned
bootstrap/javascript/python to generate markov chains
What's next for MakeTrumpTrumpAgain
scrape trump's twitter and add it
get enough data to use a neural network
dynamically generate drumpf statements
If you want to read all the machine generated text that we deemed acceptable to release to the public, open up your javascript console, open up main.js. Line 4599 is where the hilarity starts.
Log in or sign up for Devpost to join the conversation.