A potential solution to the Home Credit Kaggle competition based on neural risk modeling.
Unfortunately, our team was unable to complete our ambitious solution to the competition given our hectic personal schedules. We want to see someone pick up the challenge, so we're open-sourcing the idea, as well as the code we worked on. There are lots of TODOs, and the code isn't always clean, so WYSIWYG and YMMV. We hope this work inspires someone somewhere to innovate with the idea of neural risk modeling.
https://docs.google.com/document/d/1BD6ZXB6OSEE-FoJjtsitb5cR36fBKpO5K11O08kOHwU/edit?usp=sharing
It's recommended that you install PyTorch according to the directions here using your preferred package manager. There's a requirements.txt file you can check out, but unfortunately using PyTorch means that you likely won't be able to install everything with that (since PyTorch has prebuilt binaries for each platform).
- python 3.6
- torch 0.4.0
- pytest 3.6.3
More will be added to this list as time goes on. Please take dependency complexity into consideration when choosing which tools to use for your development and analysis! If there's an existing dependency that covers what you're trying to do, consider the tradeoffs involved with using that tool instead.
Jason Mancuso (@jvmancuso)
Barton Rhodes (@bmorphism)
Ted Troxell (@ttroxell)
Yizhe Xu (@yizhexu)
Aleena Notary (@notarya)
Lindsey Eggleston (@leggleston)
We'd like to thank the following friends and colleagues, who contributed to planning and discussion calls:
Addison Klinke (@addisonklinke)
Mark Sliva (@marksliva)
Andrew Plassard
Michal Wojczulis (@miwojc)