I’ve updated http://www.vrc-data-analysis.com with this season’s stats. I plan to update the content at least weekly. I’ve provided commentary here in past years, but I’m not sure that I’ll continue that, especially given the changes in the forum’s posting mechanisms and my perception of user engagement.
Good luck this season, roboteers!
It looks like you may have misconfigured something, as the site only seems to work at http://vrc-data-analysis.com without the “www.”
Looks like it’s the www subdomain that doesn’t work. http://vrc-data-analysis.com works for me.
Yep - missed a DNS entry…both should be working now
Great analysis tool. I have a question about how the mu is calculated. I know it looks at win and losses and skills scores, but what is the formula for this?
It does not look at skills scores, just wins/losses/ties based on match result (e.g. score - it does not factor in DQs for several reasons).
Here is the paper: TrueSkill(TM): A Bayesian Skill Rating System - Microsoft Research
Here is the implementation I use: GitHub - sublee/trueskill: An implementation of the TrueSkill rating system for Python
I don’t know how math heavy you want to get. This one totally geeks out on the math - https://www.moserware.com/assets/computing-your-skill/The%20Math%20Behind%20TrueSkill.pdf and an associated blog post Computing Your Skill (Note: I am not the author)
Also, I realize above just wins/losses/ties based on match result (e.g. score - it does not factor in DQs for several reasons) could be misconstrued. TrueSkill does not factor in the score of the match (e.g. winning 100-99 is the same as winning 100-1). I just meant that the W/L/T is based off of the match scores (which don’t provide DQ status) and not the rankings (which factor in DQ, but not at a per-match level)
So, if you win 100-99 but are DQd, you get credit for the win in my TrueSkill implementation. If you lose 100-99 but your opponent DQs, my TrueSkill implementation treats that as a loss for you.
