Our inspiration was an article that one of our team members read and then suggested to us. After reading it, we thought that the idea of automating tinder, inspired in part by the success that the writer had with his project, would be a fun project.
What it does
Our app uses a sophisticated deep learning algorithm to make a more compatible version of the popular app (celebrity) tinder for every user. This app is a single web page application that, when loaded, immediately shows an image of a celebrity acquired from the microsoft celebrity dataset and two buttons (like and dislike). By training based on the people that the user likes or dislike, the app is able to show a more personalized selection of options to choose from for the user.
How we built it
We had 2 methods implemented in our app to train the program. First, we implemented a stochastic gradient descent or online machine learning algorithm to partially train the program. We also used the deep learning library, openface, to train our program. We had ReactJS as our front end framework and implemented Django Rest on the back end.
Challenges we ran into
We had many docker setup and compatibility issues; we continued to use it, however, because openface is most easily implemented alongside docker. Lack of experience about deep learning which resulted in having to spend a lot of time reading documentation. Debugging the code was a very long and hard process; we were thinking about writing tests but because of the finite time period, we decided it would just be better to debug as we go.
Accomplishments that we're proud of
Avin is proud of the clean and smooth User interface. Eli is proud that we were able to finish the app. Hairou is most proud that it works. Kevin is most proud that we were able to have such a great team with great people. Anushka is proud that our team didnt give up
What we learned
Deep Learning is hard. UPenn students are smart.
What's next for WingMan