Sometimes I want to relax and enjoy playing against bots. Unfortunately, the reality is quite the opposite.
I’ve been trying to play against 7-9 kyu bots for many games and the results are disaster. You can see my profile: パッサコーンの碁 [11k]. I used to be around 7-8 kyu but now I’m down to 11 kyu after unbelievably consecutive losses.
I believe some bots are much stronger than their ranks and my rank is definitely not 11 kyu. And since my rank is too low and it should be, I don’t want to play against anyone. Please review and reassess the bots ranks… as well as mine.
To be clear, if you are expecting OGS admins to manually change your or the bot’s ranks, this isn’t going to happen. Ranks are averages based on many games against varied opponents. You might be particularly bad against these bots compared to against humans of your level, likewise humans you could easily beat might do better against the bots than you because they know bot weak points.
Play people with your 11k rank and you should win easily and go back up.
I will add my experience to the support. I think that the system really does not adequately rank the bots, in particular on the 13x13 board, since most of the games I played with them there. The problem is not even that my rank dropped from 3rd to 11th, I didn’t play any worse) The problem is that their rank has not increased by any punt and other players, especially those who are starting to beat them like a wall. For statistics, you can take games from my profile, nickname: Veter
Maybe it’s changing, maybe it’s not. Or it’s changing the wrong way. But here’s an example. here is a slice where the bot has about 75% of wins, including stronger players, and its rating is falling… ?
The essence of the problem, of course, is a little different. If a player with a low rank suddenly appears and starts defeating a strong bot, then after 5-7 games they should level up and the system should come to an agreement. That is, the more he wins the bot, the less the system should take away and add a rating.
Perhaps one of the problems is people who play a lot against bots and find a way to exploit their weaknesses. For instance suppose a bot is 10k and a human is 20k. If they only play together, and the human finds a way to win 50% of the time, then both become 15k, although the human would still play at 20k level against other humans, and 10k humans who never played the bot before would have 50% chance against it.
I hacked a little bit on the ratings repo to try to come up with a way of limiting rating changes from repeatedly playing the same player, but it wasn’t obvious to me what kind of stats to collect about recent pairings and how to tweak glicko updates to deweight repeated pairings. It would be real nice if someone who had a working understanding of glicko could share some insight about how to do this in a principled way.
There’s probably some naive approach that could work - like automatically make games between a pair unranked if there have already been N ranked games between the pair in some time window? Or would that break glicko in some way?
(from the point of ranking-system-confidence).
