Oh no I was referring to Master Potato leaving. Whoops.

I've got too many games atm, so, goodbye:

set_limit 1v1 Ryzys 0

add_team 2v2 "We've haven't decided on a name" 6146168723 8230371260 1

 add_team 1v1 Tjoex 4026546725 2 

confirm_team 2v2 "We've haven't decided on a name"

How often does the sheet update?

About once a day, since it's run locally. There's currently two parts:

- a script that parses forum posts and updates the sheet when you set your limit, etc. (run locally from my computer) (matter of seconds)

- me going through, updating results, and creating games (currently takes 10-20 min, which I don't get until a bit late at night)

The end-goal is a single script hosted on a server tied to a Google Service Account that anyone could add to their own league spreadsheet (with a few commands- this does come with a lot of protocol) that tell it what exactly to do. Essentially, it's like CLOT for spreadsheet leagues- i.e., a version that's easier to adopt and scale.

Basically, you make a spreadsheet league (like this, Clan League, Pokemon league, anything), give edit access to a Google Service Account (basically a funny-looking e-mail address), fill out a form telling it what to do (or it'll just assume some default values I'll figure out later), and have it run/update your spreadsheet league 4 times a day + handle user input through a linked forum thread (like this) that you can specify on the sheet.

Here's what the development timeline looks like:

- script gains the ability to update results (using my credentials); at this point, it would be hosted on Google App Engine (so add a bit of debugging here as knyte finds out everything that doesn't work)- this would be done using the query game API (I've already written a wrapper library for the Warlight APIs)

- script is able to create pairings (this one's going to be a bit tricky to do and would be the first command to really require the Controls sheet in order to function; everything before this just needs the Controls sheet for edge cases); this part would be developed separately and run locally since a mix of manual and automatic access is inevitably going to result in timing issues that I don't want to explore just yet; there's going to be lots of issues to sort out when it comes to people also having edit access to the sheet, since they could easily fuck up and create massive bugs that would destroy their league (although obviously a series of best practices could be provided in documentation to avoid these issues)

- finally, the script is able to create games; this will likely be done through the create game API and- unless Fizzer makes it less restrictive- will require me to kick out a few people from IPL and move them over to another league.

- after this, the script's capabilities will be gradually extended until it's able to handle more rating algorithms than just plain Elo, more pairing methods, etc.; admin/mod rights for leagues would also be more developed

The initial plan was to actually do all the automation first and release IPL alongside the other leagues (Infinity Learner League, which creates 2v2 teams by pairing Mentors- players with higher TrueSkill/Elo ratings- and Mentees- players with lower TrueSkill/Elo ratings; Infinity Open League, which would be IPL but with just the standard Strat templates so it's playable by lower-level accounts; Infinity Bounty League, which would explore edge cases by building a bounty hunting metagame on top of WarLight; and Infinity Challenge League, which would be a king-of-the-hill league across the same templates as IPL and help explore more edge cases). But while developing the automation tool, I made the decision to take out any extra authentication layers (e.g., the ones you use for CLOTs) and instead prevent fraud by tying a forum thread to leagues; this obviously leads to some complications in user experience- as you can see, a large majority of the people in this thread had a tough time figuring out the syntax on the first sheet- and I wanted to figure those out and get the forum script working for cases like that to require the least owner intervention.

And that's why this league is pretty slow if you finish all of your MD games in real-time (the only thing you can do about that for now is raise your Limit, but that comes with some fun side-effects).

Edited 1/22/2016 17:34:53

Eh, happens to all of us at some point or another.

add_team 3v3 Apex 635295275 314032996 816809922 1

confirm_team 3v3 Apex

confirm_team 3v3 Apex

Updates:

Expanded the secondary point system a bit:

- Secondary points (the ones used for aggregate player and clan rankings) are based on how many standard deviations you're away from the mean among ranked players in your league (non-ranked players don't get secondary points). If you're less than 1 standard deviation below the mean, you're contributing points (there's a curve of 1 standard deviation)- so, in the long run, about 84% of all players should have secondary points (so spamming isn't a very viable strategy while actual participation is). This also gets rid of the huge bias toward the top that the exponential decay function.

- Secondary points are now equalized across leagues- 1v1, 2v2, and 3v3 should have exactly the same impact on your clan rating and should be fairly balanced for personal aggregate ratings

- Right now, it's a bit early in the league, so we're only getting these aggregate scores from the 1v1 league; however, I think this will turn out to be a viable system in the future since players and clans basically get rated on their skill relative to each league they participate in. With enough participation and activity, we should be able to generate fairly accurate skill ratings (across the diverse range of strategic templates) this way.

- For aggregate player rankings, you will have a minimum of 10 points if you're ranked (the rest of the league will be lifted with you) just because negative points make things icky

Rule change:

- You can no longer participate in multiple team within a league

Some data so far:

- We're getting point results in from 1v1 for both players and clans, and it looks like we've got a break in the skill spectrum:



There's a cliff between Timinator and me but after that it flows fairly smoothly.

- Clan ratings seem to make sense so far, although with some aberrations due to the small dataset:



MASTER Clan leading the pack by a big margin as they hold the top 2 players in 1v1 (although Icelandic Turtles might take their spot soon, thanks to Buns157).

- We also track participation- nice job, CORP, for outshining everyone else here:



I think we're through with the major tweaks to aggregate ratings unless something weird comes up. I like the standard deviation-based system since it's the most logical way to compare players across leagues (since leagues might be tighter/looser than one another in terms of player competitive ability).

So how many players in the league? 49?

The template stats are interesting, especially the over and underdog thing (I'm assuming that it is just based on who is ranked higher).

.

Edited 1/23/2016 13:48:44

Let's get AWF as the top clan :D

add_team 1v1 Benjamin628 9936698896 3

@Buns157:

Oh, I should explain that (mostly for internal use). It uses a method I devised (that was fixed by Widzisz):

- Every game has an overdog or underdog, with the overdog being the player with a rating >= the rating of the underdog

- The theoretical probability that the overdog would win (calculated using Elo's formulas) is tracked for each of these games

- The outcome is tracked as well

- So we have an experimental probability for overdog wins vs. an actual probability for overdog wins; we also know the number of games/trials

- Then I just use the binomial cumulative distribution function to find out the probability that we'd have the % of overdog wins we *actually* got assuming the template behaves perfectly according to Elo's model; this is the "1-tail" p-value and I'm mainly watching for 1-tail values that get below 0.05.

- The "Bias Lean" column just tells you whether the overdog/underdog won more often than expected. If it's "None," it's exactly aligned with the theoretical calculations; if it's "Overdog," the overdog won more often than expected (keep in mind that Elo's model doesn't just see the overdog winning all the time against lower-rated players; there's a probability of upsets involved as well, with upsets having a number of causes)- and if it's "Underdog," the underdog won more often than expected.

- I don't think IPL is mature enough yet for these stats to be taken very seriously; right no they're (at best) rough guidelines and I'm working on a somewhat more distributed/democratic system to judge templates

Edited 1/23/2016 17:26:12

https://www.warlight.net/MultiPlayer?GameID=10306486
This game won't start ;/

That's cos you're opponent declined, you'll be issued a new one once the failed ones are cleared out :)

Edit: Yuret's limit is set up 0, so I'm not even sure why it issued you a game with him...

Edit2: Looks like that's an old game, it's not actually part of your limit, maybe knyte can add in a system that deletes failed games?

Edited 1/23/2016 23:11:50

Oh I forgot to delete that game. >_> Yuret's limit was set to 0 as a result of the game itself, actually.

I can't query these games so I can't delete that one except manually.

Either I kick off all players below level 46 (or something thereabouts) or just keep doing this side of the league manually.