xlvector's solution of github contest
C++
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
bagging-test get blending weight by training data Aug 7, 2009
bagging
data
include
knni-all
knni-page sources Aug 30, 2009
knni
knnu-all
knnu some of source codes for github contest Aug 14, 2009
knnui
language
popular
repo_owner repo owner Aug 18, 2009
repos
uiknn
README.textile
result-train.txt
results.txt
results0.txt top20 Aug 1, 2009

README.textile

Introduction

I am XiangLiang from China. Here is my homepage
You can follow me in the Twitter, my account is “xlvector”

Algorithms

1. knni : item based knn
2. knnu : user based knn
3. knnui : hybrid item-user based knn
4. knni-all : item based knn with language and reponame information
5. knnu-all : user based knn with language and reponame information
7. repos : using collaborator data (downloaded by github API)
8. bagging : blending different algos

9. In bagging/main.cpp, there are many pre-processing methods used
1) diversity : we find it is very hard to predict recommendations for users who watches a lot of repos. So, making diversity recommendations can improve prodiction accuracy of these users.
2) popularity : if a users does not view popular repos before, it means he may not like popular repos, so, we can delete popular repos in recommendation list.

License

Before the contest is ended, you can not use my results.txt, however, it is free for you too use my source file.

After the contest is ended, you can use everything I pushed to github.

….