trumpr

We get it. The Onion has been rehashing the same jokes over and over again, and local news organizations are not keeping up with the Joneses. Where should we best get our lousy humor from?

Naturally: politics. But the 24-hour news cycle is simply not spouting enough craziness from the likes of Donald Trump, Ben Carson, Hillary Clinton, and other presidential wannabes. So we sought to find a better way. A way that would make Donald Trump say, "Keep these computers out of my country!"

This application automates the spouting of political junk so that politicians need no longer spend time inventing and inundating the self-referential media circuit we call the 4th branch of government. Because what is democracy if not ever increasingly less human and more computerized?

History

This code was a project from HackPrinceton Fall 2015, originally built using Meteor, React, and Python on a Linode server. It has since been rewritten to be fully deployed on the standard Meteor stack.

This hack won the Best Domain Award (trumpr.org) at HackPrinceton. Thanks, Domain.com for believing in us! :')

Running

First make sure you have Meteor installed.

$ curl https://install.meteor.com/ | sh

Then clone away. Trumpr should run with no further setup.

$ git clone https://github.com/casey-chow/trumpr.git
$ meteor

Lessons Learned

This project proved interesting both to build and to see running. We originally hacked off of two languages, Javascript and Python, and while this let us quickly create the initial components for the app, it turned out integrating them was quite difficult, and so on rewrite the app was consilated into Meteor's stack.

The idea of a Markov chain came in handy here as much of the logic of this app comes from Markov chain computations. One thing that proved challenging was taking the generated text and turning them into tweets, whereas a normal Markov model demonstrationd isn't really constrained by format! Our solution was to use some rudimentary indicators of sentence beginnings and ends, and to encode the ends of tweets into the model itself. This proved useful as the generated output actually told is where to end the tweet!

Finally, we dealt a lot with the emerging (well, now very common) concept of a reactive user interface in web development. At first, this was a new concept and so we treated it like a classical interface system--we kept trying to feed stuff in through transitions, etc. etc. This actually slowed us down a lot during the hackathon, and we had concluded at the end that the reactive infrastructure wasn't suited to our app. In hindsight, however, we were just not intelligent enough (read: "we were in the process of learning") to realize how to work with the system. Upon rewrite, the reactive system actually proved incredibly conducive to productivity, shaving literally hours on debugging data flow.