Jump to content
  • Announcements

    • Mako

      Union in Chains Pictures   10/14/2017

      With union in chains in full swing, everyone wants to upload a few pictures with their battle report (they're highly recommended to ensure your report is accepted as valid). Check the Union in Chains section for some handy advice (and user avatars) before you start!  
RedSam

Guild vs. Guild stats

Recommended Posts

Recently, I've been developing a ranking system for Guild Ball that works by recording the winner and loser of each game, rather than the placing of players at the end of each tournament. This style of ranking does a few things:

  • Removes game score from determining ranking - a win is a win, a loss is a loss
  • Incorporates strength of opponents - a win against a high-ranked player should be worth more than a win against a low-ranked player
  • Allows non-tournament games to be included in the rating and ranking system (but can be filtered out if desired)

The ranking algorithm I settled on is based on the one used by the NCAA and uses win percentage, opponent's win percentage, and opponent's opponent's win percentage. Each player is given a rating that's intended to reflect not only rate or number of success but also the difficulty of his or her games.

A secondary benefit of using each game for ranking rather than only tournament results is that the guild-versus-guild matchups can be recorded. This allows the win percentage of each guild to be compared both in total and against each other guild individually.

Take a look at the player rating page and the guild stats page to see the current rankings, which include every event on TieBreak (as of about an hour ago) that includes game-by-game results since January first.

Some interesting things I noticed in the guild rankings:

  • Hunters are no longer in last place
  • Alchemists don't have a less than 50% win-rate against any guild
  • Fishermen have a greater than 50% win-rate against every guild except Alchemists
  • Fishermen are being played about 20% more than any other guild

Share this post


Link to post
Share on other sites

Some of these stats are really interesting. i had no idea Engineers had such catastrophically low win rates vs certain guilds. I always rated them pretty high. Butchers and morticians still a force to be reckoned with. Morts i understand, i really think obulus is still primed for this meta. Morts having the single strongest WR vs Alchs seems to evidence this. Butchers are surprising.

very interesting to see Union carries an overwhelming advantage vs every other guild except alchs, fish and butchers. they average high 60% WR. the only thing controlling them is alchs and fish. O_O. Not even Alchs are averaging as high win rates vs other teams that union pulls off. 

Glad to see hunters climbing though, skatha is a real help. These are competitive tournament win %s, so in casual games they should be pretty viable if they can match engis and masons for WR, two guilds i think are completely viable also.

it wouldn't surprise me to see Union rise even higher after benediction releases. vRage is going to absolutely love him

 

Share this post


Link to post
Share on other sites
19 minutes ago, HuggyTheBear said:

Is there a reason why you haven't used what has been used in the Chess-scene for many many years: elo-rating?

I did look at that, but it requires the ratings to be re-calculated after every game. Because of that, the order of games matters (i.e. if you beat player A then lose to player B, player B will get a bigger boost than if you lose to player B and then beat player A, even if both games are on the same day). That seemed a bit unrealistic to me and makes the programming and calculations notably more complex.

The Elo system also relies on a number of subjective variables, including the K-factor which is assigned to a player depending on number of rated games played, highest rating achieved, and age. These values might be well-established for chess, but it would probably take a lot of time and study to make sure they accurately reflect Guild Ball.

Also, since Guild Ball represents the playing of a sport, I thought it fitting to use a ranking system used in competitive sports. :)

Share this post


Link to post
Share on other sites

Yeah I'm much higher than BO!

Also agree it's a bit biased that all the US players have their states, yet people in other Scotland, Wales an Northern Ireland bear the insult of being labelled English. BTW I am English, but have respect for the cultural identity of the other countries within the UK.

Besides that, really great site well done and thanks for putting your time into it!

Share this post


Link to post
Share on other sites
19 minutes ago, FearLord said:

I like this system. It's ranked me 7th instead of 9th. ?

For a while, it had VanValzah right above  Nach. Pat thanked me in person :P

 

1 minute ago, CurlyPaul said:

Yeah I'm much higher than BO!

Also agree it's a bit biased that all the US players have their states, yet people in other Scotland, Wales an Northern island bear the insult of being labelled English. BTW I am English, but have respect for the cultural identity of the other countries within the UK.

Besides that, really great site well done and thanks for putting your time into it!

The system is built to accept Scottish, Welsh, and Northern Irish people as well. When I enter data, it saves the region for each new player as the region of their first reported tournament. So far, there's only been one event on TieBreak in the U.K. that wasn't in England (as far as I can tell, at least). If it's any consolation, I'm sure a lot of the US players are in the wrong states :)

I'm happy to change the home location of any player - just shoot me a message.

Share this post


Link to post
Share on other sites
20 minutes ago, RedSam said:

For a while, it had VanValzah right above  Nach. Pat thanked me in person :P

 

The system is built to accept Scottish, Welsh, and Northern Irish people as well. When I enter data, it saves the region for each new player as the region of their first reported tournament. So far, there's only been one event on TieBreak in the U.K. that wasn't in England (as far as I can tell, at least). If it's any consolation, I'm sure a lot of the US players are in the wrong states :)

I'm happy to change the home location of any player - just shoot me a message.

I just noticed I spelled Ireland wrong! Another insult! 

Makes sense what you say about the location and surprising that there are not more events in the rest of the UK tbh.

Share this post


Link to post
Share on other sites

Well done. While I still have no opinion on the player ranking thing vs BO, the guild ranking stuff is good to see. Would be interesting to see the seasons final stats archived and reset the counter for future seasons. 

Best part is being able to compare guild vs each guild instead of a nebulous overall win%. 

Share this post


Link to post
Share on other sites
5 minutes ago, Frostmane said:

You should put your code on github ;)

I'd be happy to share my code, but it doesn't really do much without the database of player and game data. Is there a specific purpose you'd want it shared for?

Share this post


Link to post
Share on other sites
4 minutes ago, RedSam said:

I'd be happy to share my code, but it doesn't really do much without the database of player and game data. Is there a specific purpose you'd want it shared for?

Posterity ;)

Also I'm mostly just curious what you are doing to pull the data in. I was going to do basically exactly the same thing but never got the time

Share this post


Link to post
Share on other sites

One hopefully minor improvement could be done on the mobile site. The banner at the top of the page where you can navigate the site blocks the 3 buttons at the top which let's you change to tournament only for example. You can still click it on mobile if you aim low, but I didn't even see them at first.

Share this post


Link to post
Share on other sites
1 minute ago, Frostmane said:

Posterity ;)

Also I'm mostly just curious what you are doing to pull the data in. I was going to do basically exactly the same thing but never got the time

I'm just copy/pasting it from TieBreak. I made a scraper that lets me do whole rounds of a tournament at once, so it only takes a couple of minutes to enter each event.

You can also enter rounds manually, but it takes longer and requires you to enter players ahead of time. This feature is open to anyone and was intended for people to enter casual games.

2 minutes ago, Ik-tornado said:

One hopefully minor improvement could be done on the mobile site. The banner at the top of the page where you can navigate the site blocks the 3 buttons at the top which let's you change to tournament only for example. You can still click it on mobile if you aim low, but I didn't even see them at first.

Good catch! I made some changes to how things load the other day and forgot to change the top of page buffer. I'll fix that today.

Share this post


Link to post
Share on other sites
39 minutes ago, MarkM said:

How does your system get someone with a 3W/12L record almost 100pts above some one with an 8W/2L record?

Probably the same way that results in #2 having almost twice the number of wins and games played as #1, but the same number of losses - but one less tie.

You could say he was...... broken by that tie.... *yeaaaaaaaaaaaaaaaaaaahhhhhhhhhhhhhhhhhhhhhhh*

Share this post


Link to post
Share on other sites

I suppose I should chime in on this...

Firstly, I always said that I'd be happy to see someone else coming up with their own rankings system. Now that it's finally happened I'm not sure how I feel. Certainly I have nothing but respect for @RedSam for the work he's done on. But half of me can't help but worry that I'm going to become redundant :'-(.

That all said, there's plenty here that is superior to what my rankings has to offer - the match up data in particular is very interesting. I always said I'd love to be able to do my rankings based on similar data but always thought it impractical to gather. It's nice to be proven wrong about this. 

That said I think there's still some things that my rankings does that this doesn't (yet?). Mine does go back over previous events to work out a weighting for the event based on previous performance, and includes data from events that haven't been input into tiebreak.

Whilst this second factor is becoming less and less important, I think with this data we could start to create something more like an ELO rankings (there's plenty of ranking systems outside of chess we can look to), and I'd love to be able to get my hands on this dataset to see what I can do with it. Maybe there's the chance for @RedSam and myself to work together to try and get team best of both systems? 

Cheerio,

Ben

Share this post


Link to post
Share on other sites
8 minutes ago, malladin.ben said:

I suppose I should chime in on this...

Firstly, I always said that I'd be happy to see someone else coming up with their own rankings system. Now that it's finally happened I'm not sure how I feel. Certainly I have nothing but respect for @RedSam for the work he's done on. But half of me can't help but worry that I'm going to become redundant :'-(.

That all said, there's plenty here that is superior to what my rankings has to offer - the match up data in particular is very interesting. I always said I'd love to be able to do my rankings based on similar data but always thought it impractical to gather. It's nice to be proven wrong about this. 

That said I think there's still some things that my rankings does that this doesn't (yet?). Mine does go back over previous events to work out a weighting for the event based on previous performance, and includes data from events that haven't been input into tiebreak.

Whilst this second factor is becoming less and less important, I think with this data we could start to create something more like an ELO rankings (there's plenty of ranking systems outside of chess we can look to), and I'd love to be able to get my hands on this dataset to see what I can do with it. Maybe there's the chance for @RedSam and myself to work together to try and get team best of both systems? 

Cheerio,

Ben

I think it's useful to have rankings that look at more tournaments other than tiebreak - this data ignores one of my tournament wins from this year for example, because it wasn't run through tiebreak. There's pros and cons to any system really, but consistency is key. 

Share this post


Link to post
Share on other sites
4 minutes ago, malladin.ben said:

That said I think there's still some things that my rankings does that this doesn't (yet?). Mine does go back over previous events to work out a weighting for the event based on previous performance, and includes data from events that haven't been input into tiebreak.

I'd be interested to learn more about the nuts and bolts of how your system works. I have noticed that you have results that aren't on TieBreak, which is definitely a good thing. My system is definitely capable of taking in other events (and even one-off casual games, local leagues, etc.), but it would take people volunteering to submit those data.

Whilst this second factor is becoming less and less important, I think with this data we could start to create something more like an ELO rankings (there's plenty of ranking systems outside of chess we can look to), and I'd love to be able to get my hands on this dataset to see what I can do with it. Maybe there's the chance for @RedSam and myself to work together to try and get team best of both systems?

I looked into several different ranking systems when I was considering what to use. Since I was trying to distance this as much as possible from the current tournament tie-break system (and thus avoid that discussion altogether), the RPI system seemed the most robust. It also benefits from not requiring large amounts of data. I originally built this for handling the results from a local league, so I wanted something that could give decent rankings based on only a few games per player. As more games are entered, the results in my system should converge with the results in yours. Central limit theorem and all that.

I'd be more than happy to work with you! One thing that could really improve my system would be a more automatic data-entry method - I've made it pretty streamlined, but my system still requires manual entry of each event. It would be really great to have some kind of data crawler that could pull new events from TieBreak as they happen...

2 minutes ago, FearLord said:

I think it's useful to have rankings that look at more tournaments other than tiebreak - this data ignores one of my tournament wins from this year for example, because it wasn't run through tiebreak. There's pros and cons to any system really, but consistency is key.

One of my greatest laments is the missing data from the Las Vegas Open. Any ranking system gets better when you incorporate events with people from multiple regions. Missing one of the biggest US tournaments this year, particularly one that featured players from all over the country, is unfortunate.

That being said, I'd be thrilled to add non-TieBreak events to my system if anyone who has those results wants to send them to me.

Share this post


Link to post
Share on other sites

Someone contacted me about the possibility of a screen scraper, but I'm. Not sure who it was and can't find the message anymore. I was intending to have a look into it over the summer when (being a teacher) I have some time to kill. I'm right at my busiest time if the year currently, so can't do anything fit about a month, but should be able to start rolling some balls by the end of may

Share this post


Link to post
Share on other sites
2 hours ago, MarkM said:

How does your system get someone with a 3W/12L record almost 100pts above some one with an 8W/2L record?

A couple of reasons:

  • There is a factor in the rating system based on number of games played, and fifteen games is above the cusp (that player is getting about 96% of their actual rating) while the other is right on the edge (probably only getting about 60%, as the 50% cutoff is at around 10 games right now). I had to build this in because there are people who have only a few games in the system, but those games are all wins against high-ranking opponents. For example, without that factor, the list is dominated by someone whose record is 1-0-0 and that one win is against @Slothrop. A good win, to be sure, but I don't think having that person ranked #1 reflects reality. (I posted the raw RPI ratings here if you want to see how different they are.)
  • The system also incorporates strength of opponents. While I haven't looked into those two players' game history, I would guess that the 3-12 player has played against a set of higher-ranked opponents than the 8-2 player. While it does occasionally lead to weird rating disparities (as is the case here), this is an intended feature of the system. It is designed to consider not just wins and losses, but who the wins and losses are against.

Share this post


Link to post
Share on other sites

FYI I talked with the dev from Tiebreak a couple months back and thand renwere working on a REST API so it's possible that screen scraping is a thing of the past soon(ish). I can probably find his twitter handle if I look through my message history and filter out the fighting with Australians

Share this post


Link to post
Share on other sites

Quick question.

 

How does it handle the issue of tiebreak scoring games as a tie when it comes to witness me and scores of over 12pts, which at present tiebreak cant handle?

So you have me down as 15 played, 13 win, 1L 1T in #1 spot.

My Tie is shown as against henry kay, when in fact it was a win to me as we both got to 12 due to his witness me and i then killed a player to go to 14. It was in an event i won. 

But due to tiebreak limitations and the way it scores for further tiebreaks it shows it as 12

My score is actually 15played 14win 1L

Apart from that i really like that it takes into account the quality of players you play to rank you which gets around easier rides to high place finishes and also tiebreak dropping you to 6th after losing in the final

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now


×