[22] Swipe Smart

Inspiration

For guys, finding a match on Tinder is largely limited by your amount of swipes. This project automates that process based on your preferences.

What it does

Swipe Smart makes a custom dataset of like/dislike preferences, trains a deep learning model, and then swipes on Tinder for you based on the model's predictions.

How we built it

Swipe Smart is made from a series of components: 1) scraper.py - scrapes tinder for nearby user photos 2) label.py - lets you manually sort photos into like/dislike using a simple GUI 3) preprocessing.py - converts all photos to greyscale and resizes based on detected faces 4) train.py - trains a modified version of VGG19 to predict your preferences 5) bot.py - uses Tinder API to swipe based on the model trained

Challenges we ran into

We weren't able to properly debug train.py because of lack of resources on laptop, should have used cloud to train instead.

Various smaller considerations about creating large enough dataset given not very much time, avoiding possible bias from our sources.

Additionally, pynder crashes due to an issue with init on its User objects, which we had to modify (so naively installing pynder will not work).

Accomplishments that we're proud of

It's a heck of a lot of work for 2 people

What we learned

Tensorflow and Python3 practice, with some OpenCV

What's next for Swipe Smart

QOL changes, fixing train.py (and evaluating different image classification models), changing scraping methods, building bigger/better datasets faster

Built With

• opencv
• pynder
• python
• tensorflow