PDA

View Full Version : Official rules of the tournament



Kletian999
04-07-2013, 08:29 PM
Having completed my 2nd tournament (top 20, woo) it occurred to me that there's a lot of the tournament that isn't explained anywhere in game. While I've come to understand most of it I feel it's helpful to have it all written down somewhere. I'm hoping more experienced folk can correct or fill in where my understanding runs short.

General: the tournament games are locked to a specific team level, and are a separate queue from quick/rank matching. Thus you will only fight someone looking for a tournament fight the same time as you. There's a small cost to enter and a large reward for finishing in the top 3. You are only allowed to seek 5 battles a day (25 total) and the day switchover is somewhere in the afternoon, Eastern Standard Time zone, (ends Sunday around dinnertime EST. If you hit the "find a new game" button after your 5th battle that day... what happens? Battles not taken on early days can be played on later days. The winner of the tournament is determined by the highest Tournament specific ELO.

ELO
Each Tournament is considered a fresh start to Elo rating (1000), meaning matchmaking on the first two days you never know who you'd encounter. Just like normal, the gains and loss of ELO is influenced by the Elo of your opponent. ~15 points for an even fight, going as little as 2 or as high as ??? on a very mismatched fight. While your overall Elo doesn't influence the matchmaking, it will be modified by your wins and loses as tournament matches are basically ranked matches.

Going by this, it seems optimal strategy is to wait until the last days of the tournament, and play when few but higher Elo people are playing. The gain/loss imbalance will favor you provided you don't lose to someone who had been beaten below 1000 beforehand (and of course, can win a 2:1 ratio or higher). Likewise, once you have a top rating, you only want to play at high volume time so your opponents will be close matched (perhaps even resetting the search every 60 seconds to keep your allowed Elo window small), to mitigate potential loss and maintain reasonable gains.

Miscellaneous
There's no requirement to stick to one team as you play the tournament, nor is completing all 25 games necessary or even advisable for a win depending on your circumstances. A battle started before the tournament end but finished after the end time does not register to the tournament. Tournament Hall of Honor statistics take a few minutes after the battle finishes to display properly.

If my understanding is correct, there's a lot of flaws in this model. From reading past posts, this tournament structure was a quick implementation since it has no bearing on the Single Player game, which is the dev priority. Thus, adjustments or alternative tournament formats are not forecasted.

sweetjer
04-08-2013, 03:37 AM
Well, I wouldn't call your ELO section a "rule" of the tournament, but it is the correct meta-strategy. I think maximum gain/loss is 32 points, but I may be wrong on that. I'm not exactly sure of your analysis of resetting the window. I suppose this levels your gains, but it might be better to attempt to snag a top level player while your are 1000 in an attempt to cause the greatest gain for you and loss for them possible. That's assuming you can win, know they're online, don't pull at match at the 2pt end of the spectrum, etc. Not sure how that ultimately pans, and I'd like to hear a tourney winner weigh-in on that particular strategy. I think a note on tiebreakers is important. In case of an elo tie, winner is chosen by total wins first and then win/loss. As I haven't participated much in tourney (runs on the worst possible days for me, though my schedule recently changed), some of my information might be a bit off. Good thread.

Aleonymous
04-08-2013, 04:22 AM
Yeah, its an interesting thread.

So, you're saying that the "tourney MM" looks for an opponent randomly (regardless of your "global Elo"?), and then applies the appropriate +/- for victory/defeat to a "local/tourney Elo". And, I assume that the +/- depends on the "global-Elo" difference of the opponents? Finally, apart from the tourney-ranking, the match outcomes affect your "global-Elo", just like Ranked Matches.

It has some subtleties to it, I agree, that leaves space for such tactics & exploits in order to best the tournament. It seems it's not as simple as "beat everyone the MM pairs you against"...

Leartes
04-08-2013, 04:57 AM
I think it doesnt look randomly, but matches according to tournament elo. If you have 1100 tourney elo it should take about half a minute to open up to +/-100.

Aleonymous
04-08-2013, 06:08 AM
I think it doesnt look randomly, but matches according to tournament elo. If you have 1100 tourney elo it should take about half a minute to open up to +/-100.

You are right; this makes more sense, so I guess that's the way MM works.

So, recap:
(1) Upon registering the tournament, you're assigned a fresh local-Elo = 1000
(2) Tourney Match-Maker pairs you with an opponent close to your local-Elo
(3a) Depending on the outcome, apply +/- to local-Elo with respect to local-Elo-difference
(3b) Depending on the outcome, apply +/- to global-Elo with respect to global-Elo-difference,
(4) Goto{(2)} or halt, if limit 5-match/day or 25-match total is hit.

Kletian999
04-08-2013, 07:12 AM
Aleo, your understanding matches mine. I presume much like normal MM, the allowed Elo range can grow to quite large if left open without a match. This week I routinely waited 5+ minutes for a tournament battle off the weekends.

netnazgul
04-08-2013, 08:47 AM
Concerning Elo change - there is a maximum of 32 and minimum of 0 (dunno whether it was fixed to 1 min, generally it was down to 0, especially dreadfull for Tirean in ranked sometimes :) ). Even matches give/drain 16 Elo for each player. Effective score is calculated using the logistic curve, so it's pretty steep and at 100 elo difference you'll only win ~12 points and lose 20, with 200 elo difference it's 8 and 24 and so on.

The bad side of coming late into tournament is that you need to play more matches in less time and can get into situation where there are no high-elo players online currently (cause they have played their today's matches already) and you wait for 20+ minutes in a queue to get a match (occured to me on saturday around 12:00 UTC+0). So common playtime should also be considered - in most cases it's the first 2-3 hours after new bunch of matches is given. Still you'd normally enter the tournament at least 24 hours after it started, as it can get you some points boost to play vs more rated opponents (so you can achieve more than 16 points in your first matches as you would have normally vs even opponents).

To snipe your opponents in tournament you can use Steam Friends, as you are notified when someone in your friends list is logged on/off in Factions.


All in all this tournament system has a lot of flaws in it, that is what I can agree upon. But nevertheless it's still quite objective on players' skill and current form (especially the latter) as you need to maintain a close to perfect W/L ratio (so far you needed 100% win ratio to get the first place, this changed with the latest tournament as everyone had at least 1 loss on his scores) and be able to play good against skillful players whom you will certainly be matched against at least several times.

Aleonymous
04-08-2013, 12:09 PM
I'm posting a figure depicting the ELO-change curves with respect to the ELO-difference between the players (+/-500 range) and to the match outcome (victory/defeat). A K-factor of 32 is used, i.e. assuming both opponents' ELO ratings are below 2100 (?).

https://dl.dropbox.com/u/99633386/TBSF_ELO_Change_Kfactor32.png

It validates netnazgul's benchmark values at Δ{ELO}=0,100,200 etc

Kletian999
04-08-2013, 01:02 PM
In the tournament, I don't think I've ever seen a local ELO >1300 so that makes sense.

netnazgul
04-08-2013, 02:24 PM
You can search somewhere on beta forum for the function John uses to calculate Elo. K-factor is 32 for below 2100 and is decreasing [linearly] from 2100 to 2400 becoming 16 there.

Kletian999
04-08-2013, 02:49 PM
I glanced at the wiki page once, what does K factor mean in layman's terms- biggest point changes or lower point changes?

Fun fact, looking at Slimsy's stream of my battles with him over the weekend, he actually fought me using regular ranked match for our rematch, while I was in tournament queue...

Aleonymous
04-08-2013, 03:16 PM
You can search somewhere on beta forum for the function John uses to calculate Elo. K-factor is 32 for below 2100 and is decreasing [linearly] from 2100 to 2400 becoming 16 there.

If I'm getting this right, K-factor is the number of ELO-points exchanged in each match. E.g. at equal rankings, a win/loss => +16/-16 to each player, amounting to a sum of |16|+|-16|=32. The reduced K-factor for high-ELO matches, essentially means that ELO-ratings of those grand-masters are less fluctuating, right?

Here is the thread of the beta-section discussing the Ranking/ELO. (http://stoicstudio.com/forum/showthread.php?512-Ranking-ELO)

The formula is pretty simple, assuming a constant K=32 for simplicity. Pseudo-code following ...


// Inputs :
e1 == ELO of player#1
e2 == ELO of player#2
sc1 == Score of the match for player#1 //sc1={1,0}={win,loss}, and sc1=0.5 for a draw (n/a!)
// Calculations:
sc2=1-sc1; // Score for player#2, so that sc1+sc2=1
Q1 = 10^(e1/400);
Q2 = 10^(e2/400);
sc_est1 = Q1/(Q1+Q2); // Score-estimate for player#1, i.e. ~1=win when Q1>>Q2
sc_est2 = Q2/(Q1+Q2); // Score-estimate for player#2, i.e. ~1=win when Q2>>Q1
delta_elo1 = K*( sc1 - sc_est1 );
delta_elo2 = K*( sc2 - sc_est2 );

franknarf
04-08-2013, 03:34 PM
Looking at your graph, the K-factors is the vertical gap between those curves. Much less than 32 can be exchanged if (i) one player is the strong favorite to win and (ii) that player does win. E.g., against a player with 200 higher ELO, I'm expected to win ~1/4th of the time, so...

if I win, I take 3/4*32 = 24 < 32 points from my opponent;
and if I lose, they take 1/4*32 = 8 < 32 points from me.

Either way, the points exchanged are less than 32 for any uneven match.

Maybe in layman's terms it's the size of the pot (think about betting in a card game or something), with the stronger player having to put a larger share of the K=32 into the pot (because they are expected to win it).

Aleonymous
04-08-2013, 05:38 PM
Looking at your graph, the K-factors is the vertical gap between those curves. Much less than 32 can be exchanged if (i) one player is the strong favorite to win and (ii) that player does win.

You're right. Actually, if your criteria (i)+(ii) are met then its always less than K elo exchanged; and, if (i)+NOT{ii} are met, then its always more than K exchanged. When NOT{i}, then the points exchanged are close to K.

What is a bummer is playing a RM, barely winning, and finding out that the opponent was QMing... So its just +2.

EDIT (after franknarf's clarifications!): K is the max of ELO exchanged. That happens when the players are evenly-matched or when something really unlikely happens (e.g. a very weak player wins). In all other instances, the ELO exchanged is lower than K.

franknarf
04-08-2013, 06:02 PM
Oh, hmm. Just replying to that part about "more than K exchanged": I was/am under the impression that K is the maximum that is ever transferred. In an evenly matched game, 16 will change hands. In an uneven match, either the favorite wins, so little is exchanged; or he/she loses, so up to (but not more than) K are exchanged.

The amount exchanged is 32*(1-probability that winner was expected to win given the players' Elo difference) <= 32...? In the code above, the deltas are of this form.

Aleonymous
04-09-2013, 01:38 AM
Oh, hmm. Just replying to that part about "more than K exchanged": I was/am under the impression that K is the maximum that is ever transferred.

Once again, you're right. Late-night posting in math-related threads should be avoided! :cool:

(I am correcting/edit my earlier post)