Thread history
Viewing a history listing
Rumble Pairing Selection Algorithm
Beaming raises concern over a ranking of a bot in the roborumble and the number of matches it has played with certain opponents. Xor clarifies that LiteRumble selects missing pairings first and each bot should have a similar number of battles given enough time. GrubbmGait speculates that there may have been a hiccup in the rankings last year. Xor suggests that older bots with more than 5000 battles aren't selected until everyone has had at least 5000 battles, and newer bots have a priority because they have less than 5000 battles. They also suggest that the current algorithm doesn't have any problems and that older bots will eventually get more battles in the next 100 years. Beaming thanks Xor for the explanation.
Hi, I just looked at ranking of Vodious Jen http://literumble.appspot.com/BotDetails?game=roborumble&name=voidious.micro.Jen%201.11&order=-Battles
for a bot which is in the rumble for many years, there are quite a lot of opponents with which this bot has less than 5 matches and those are bots which are also in the rumble since forever.
I wonder why match selecting algorithm avoids particular bots. Is it normal?
What do you mean by avoiding particular bots. From the code, LiteRumble selects missing pairings first, then pairings with less battles, etc.
Given enough time, each bot should have similar amount of battles.
Ups, I posted wrong link to Jen rating, see http://literumble.appspot.com/BotDetails?game=roborumble&name=voidious.micro.Jen%201.11&order=-Battles
I do use LiteRumble to check for my bot improvement (if I run clients anyway, why not update "real" rating). But the issue I report is unrelated to my bot.
If you look at Jen pairing, you will see almost 200 bots, with which it has only one pairing. This is not normal for a bot which is many years in the rumble. http://literumble.appspot.com/BotDetails?game=roborumble&name=voidious.micro.Jen%201.11&order=-Battles
- We need to find out whether they have only 1 pairings 10 years ago. If so, we could add back the missing pairings from backup.
- From the code, "Battles" is merely a counter. If it shows 1 pairing, then there must be data loss unless it's 1 initially.
as far as I can remember, some time ago (~1 year) there was a hickup which resulted in the strange rumblerankings. Bots had disappeared from the ranking, and when they came back after a battle, their stats against others was either complete or just 1 battle. Just can't find evidence of it on the wiki
The pairings are already missing in 2021, so it mustn't be resulted from the 2022 hickup http://web.archive.org/web/20210127174934/http://literumble.appspot.com/BotDetails?game=roborumble&name=voidious.micro.Jen%201.11
A strange thing is that most pairings wit 1 battle (of nz.jdc.nano.NeophytePattern 1.1) are either 2020.10 or 2022.3 (which is the exact time the last hiccup happened). This may indicate that the loss of pairings are resulted from restoration process.
Would it mean that there is a lingering database which has old pairing number? I.e. literumble thinks that there are plenty of pairing (thus not asking for more battles) but it shows in the stats only actual ones.
No, you can clearly see from the code that all data is saved into bot.PairingsList, which gets overridden each time a new battle is ran. So if a pairing is missing, the data may not be lost, but if a pairing shows 1 battle, then that's all LiteRumble has.
Skilgannon once said that he had backups. We could ask help from him and do a merge, e.g. take whatever pairing with max battle counter.
