Need a recommendation system? Email me: chebuctonian at gmail
Theories are tested here before they’re passed on and incorporated by my teammate Liang: github.com/xlvector/xlvector-github-contest/tree/master
Our first big increase in score came from suggesting forked repositories. To test that, I used blend_unwatched_sources.rb on a file of 20 suggestions per user.
Next up was creating a 2nd test file to ensure our algorithm wasn’t overfitting.
First I created a pre-processed file with user_watches.rb:
ruby user_watches.rb > user_watches.txt
Next, create new training_data and a hold-out set:
cd data; ruby prepare_training_set.rb
The score.rb file lets you test locally, instead of having to constantly push to the github repo.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent