Training models for machine learning is an intensive process. Training requires real people's time and energy, as well as expert knowledge. To reduce this gap, we created a platform we call Datafighter. Datafighter turns training into a game, where text and image-based classification tasks are presented to an end user.
What it does
Datafighter takes a CSV file, or any other structured data, and presents the user with the option of tagging items in that data set with relevant tags. Datafighter is composed of a front-end we call the "Data Arena" (the game) and a back-end (taking user choices from the game, and populating a data frame). For LA Hacks, we made the front-end/game portion to support the back-end we were already developing.
How I built it
This was built in Python, using GCP and MongoDB as a backend. We also chose Twilio to support our ecosystem.
Challenges I ran into
The biggest challenge was finding good data sets to present the user, so we made our own! Another challenge was separating our platform into smaller components, each of which could be considered a hack in and of itself.
Accomplishments that I'm proud of
We divided our tasks up really well and came up with several complimentary projects that coincided and supported each other.
What I learned
Data is dirty!
What's next for Datafighter
We are taking Datafighter to Ycombinator!
Try It out
flask, mongodb, numpy, pandas, python, twilio