coraallensavietta / RankingMethods

A Weighted Loss-Based Bayesian Ranking Method

0 stars 0 forks source link

Exploring weighting behavior around sample size n and closeness of p (for Monday 9/24) #13

Open coraallensavietta opened 6 years ago

coraallensavietta commented 6 years ago

[x] Move one large sample size n around and document ranking behavior
[x] March #1 slowly farther from #2. Fix everything except 1
[ ] Zero 1. What happens when we weight only #1? Only #1 and #2?

coraallensavietta commented 6 years ago

done commit 7d05698149502ab6372f00c3b793909b1df0345b

for next week:

[x] Redo spacing test but change n size at 1 or 2, make everything else deterministic
[x] Redo #1 test from last week & make epsilon larger? powers of 10? play with real data, visualization. play with rank loss, pr loss, weighting

notes from meeting in ranking_sim_debug.r:

[x] fix sample size for 1 or 2 as big. see what happens as you move
[x] this problems in this ranking is because there is an issue with posterior sample (still randomness here!) change n size here, make everything else deterministic
[ ] in rank function, tell it to use random for ties if using
[ ] play with real data, visualization. play with rank loss, pr loss, weighting (using code he sent 9/24/18)