I am thankful to have you as part of the community. Recently, I was watching old matches using the player profile lookup feature within the Yomi app. I had forgotten how many great players had come and gone during the life of this game. Frankly, it was depressing. We have lost more players than we have gained.
After some contemplation, I focused on the positive recent events. Your work and enthusiasm was the main positive contribution that I have witnessed during the last year. We have so many great tools at this time. These three are the most important to me.
Quick Match Ranking and W/L Tracker (Sirlin Games shout out)
There is only one other thing that I can think of that would be fantastic. It would be awesome to see a matchup chart filter that could view the match results between players in the top 5/25% of the ELO chart. One of the arguments against the validity of the matchup chart is that it doesnāt reflect expert players. This would put an end to that argument.
Keep in mind that this is not a request. I appreciate the work that you have done already. This is more along the lines of an appreciative and wishful thinking post.
I was actually just working on something like that. My first attempt was to basically look at only matches between players who were experts with characters, when they were playing those characters. But there just wasnāt enough data to be definitive. My next attempt was going to be ELO based, like you suggested.
Well, this is a bit embarrassing. I just caught a bug that was basically ignoring sets that were won by the second player from the Elo calculations. Iāve fixed it now! (Iām also working on answering the expert-player-matchup-chart question, @CKR. Iāll post the v2 of the Elo calculator when thatās done)
For character experts specifically, I think I was looking at people with a certain number of games played who were in the top N% of winrates for the character.
For the upcoming v2 spreadsheet, Iām planning on looking at sets played between players who have played at least 15 games, and who are both in the top 25% of active players (and separately, those who are in the top 5% of active players). Active players, in this case, means people who have played in the last year, which is the basis of the percentiles in the current Elo spreadsheet.
Upon checking both the ELO chart and the historical matchup chart, Iāve located a match of mine that was never entered into either. I lost 4-0 to snoc in losers finals way back when.
Ok, Iāve finished Yomi Historical Elo Calculator - v2. This version has matchup charts that are generated for all games played between players that were in the 75%ile or above at the time the set was played (these are labeled the Advanced Matchups). The 95%ile matches are also measured (in Master Matchups). In both sheets, the actual MU number is surrounded by a range estimated using the Wilson Score Confidence Interval (as described by Evan Miller).
EDIT: The Elo ratings should matchup between the two versions of the sheet. If you notice that they donāt, please let me know, as that would indicate a bug in one of them.
Loooks like the only two matchups that āmasterā chart has any appreciable data on are Grave/Zane and Troq/Zane. But some of the advanced numbers are quite interesting:
Grave Valerie has Grave losing 12-20
Rook Geiger has Rook up 12-8
Grave Arg is even, 21-22
Sets quince is 15-8, while Geiger Quince is only 14-11
Sets Onimaru is 22-13
Geiger Troq is 43-47
Troq BBB is 32-28
Setsuki Zane is 31-35 (but she is losing that one in the overall too)
DeGrey Zane is even 36-37
Gwen Zane is 12-9
Iām back with more Elo shenanigans. In my first pass, I was treating the set win/loss as the important number to attempt to predict. I decided (very selfishly, after winning more games than I expected to in my last tournament matches against some top players) to see what would happen if I looked at expected number of game wins instead. Elo is still computed after every set, but rather than just treating the set as 1 win, the Elo ratings are used to predict the number of wins given the number of played games.
This does dramatically alter the current estimated Elo ratings, and makes some players Elo tracks significantly swingier. The more important thing in my mind, though, is that if you compare the Validation tabs, the Game-based Validation tracks much more closely to the expected value curve predicted by the Elo ratings. In my mind, that makes the Game-based version a more accurate predictor overall, and as a result my inclination is to switch all of the calculations over to that basis.
Iāll also note here, mostly as a prod to get myself to actually fix it, that currently the validation is based on the Elo after the set is played (because thatās what is currently output), but should in fact be based on the Elo before the match is played. Updating that is going to make the spreadsheet unusable for a bit as I get all of the column headers fixed up, though, so Iām going to hold off on that for the moment.
The set win is the more important of the indicators IMO, but I know people disagree with that. Honing in on the opponentās tendencies over the set, and counterpicking them into a corner are both big parts of the job.