Posts from Player rating for tournaments across all installments of Heroes.

Thank you very much to the esteemed IronAxe and AmberSoler for your suggestions. But I have one request: could you each express your rating system in a single message? I've read to the end and already forgot what I started with. I might just give you all a piece of my mind ;).

Here is the link:
/topic/1565/

VDV_forever outlined my entire scoring system there.
It’s in the Heroes III topic; it has something about ratings right at the top, in the first post on each page.

The essence of our disagreement with AmberSoler is that in his improved but more complex formula, the component that depends solely on the player's position in the tournament has too much influence, while the component based on the number of days played is negligible and practically doesn't matter. He agrees to adjust this by making the strength of the tournament proportionally affect the ranking relative to the "days played" component. The weaker the tournament, the less impact this component should have. This is important to me. For details, contact AmberSoler.

The second point is the rate of rating decay and the grace period when a player does not receive any rating decrease for missing tournaments.
My suggestion is to reduce the rating by 1/3 but provide a couple of months of grace for any reason for missing a tournament. It is very difficult to take into account, especially for the General rating for all versions. After all, players may also miss them for reasons beyond their control - for example, they only play version 2, and the organizers haven't hosted version 2 tournaments for three months. In this case, the player is not to blame, so we shouldn't penalize their rating.
AmberSoler suggests reducing everyone's rating at all times without going into detail, to avoid complications with tracking who plays where and when (since we have versions 1, 2, 3, 4, 5), but reducing the rating by 1/10.
This point isn't critical to me, and I generally agree. However, during our discussion, I pointed out some points that don't quite align with my understanding of ranking policies using examples.

Decide what works best for you. I need to try to complete a tournament in version 4 against Zveroboy on this resource :) The rules are very original and correct, but the tournament is more like a training ground because participants can share not only impressions but also specific walkthroughs of certain moments.

And in general, I think that in the overall table, across all versions, there should be another feature: the ability to see the rating in each individual version. Of course, a player who only plays version 2 will not overtake a versatile player in the overall rating. But the overall table will show that he is a master in version 2. And the overall table is simply for convenience, so you can see everyone at once.

That’s complicated :) However, it is quite feasible if the coordinator of ratings for all versions is one person. That is, organizers send him the results of their tournaments. He searches his list for players, sees who did what. Calculates the tournament results. Enters them into, say, a common table of rankings for version 2, if they sent him the results from a version 2 tournament. Then enters these same rankings into the overall ranking table on the resource. In short, the coordinator must have all the lists, attendance records, fingerprints - which means tons of files. They need to work closely with the tournament organizers on this resource. The main files are five common ranking tables for each version and one overall summary table. And many archive files from organizers with tournament results and corresponding rating calculations for each tournament. An archive is needed for control recalculations in case of major situations. Well, how about it? Fun :)

P.S. If the rating decays by 1/3, then a master in version 2, if he plays consistently, entering the top 5 in every tournament, can easily overtake a versatile player if that player shows inconsistent results in at least one of the game versions. This is important.

Because with each calculation of results, the rating decreases. If a player plays three tournaments a month, they will guaranteed reduce their rating by 1/3 each time. But whether they earn enough rating to compensate for the loss and climb up depends on themselves and their results in each tournament. With Amber's proposed gradual reduction in ratings, cool universal players will gain 1000 points over a certain period, and they won't be able to increase their rating anymore because the leader should receive a maximum of 100 points per tournament, and decay of 1000 also amounts to 100. What will AmberSoler do in this case - he dodged the answer in our discussion. And with a decay rate of 1/3, you still have to try to accumulate a thousand points - you'll lose them quickly, even if you are a universal or a master in any of the versions.

By the way, I forgot to mention that giving the leader a max of 100 is also AmberSoler's suggestion. I had 50. However, it’s actually easier to calculate with 100, and I don't mind. But increasing the maximum rating gained per tournament while reducing the decay from 1/3 to 1/10 can lead to unpredictable consequences if some monster surprises you by accumulating 1000 points.
And you might get annoyed by Vasya Pupkins who are tired of watching slowly declining mediocre players and one-day wonders who got a good rating a couple of times and disappeared from the resource.

In short, think about it and involve people in the discussion, or maybe there's no need to involve anyone :)

Well, essentially, almost everything is stated correctly, except for one point that I will elaborate on below. So, two independent but mutually influencing ratings are to be considered.

1. Current rating. It dynamically reflects the current state of affairs, taking into account the player's actual result in the tournament and the player's activity.

Features:

1.1. The rating must undergo a mandatory periodic aging procedure. The aging coefficient determines the rate at which the rating results are updated. Obviously, the higher this coefficient, the greater the influence random factors have on the position of players in the table.

A leader’s accidental (forced) absence from a tournament (an objectively strong player) throws him down to the middle of the table at best, while a combination of negative circumstances for leaders can bring a random and not very strong player to the top spot. IronAxe’s argument – “let every player have a chance to be at the top” does not hold up because this “chance” is due not to the game itself (not to the strength of the player, as the rating should confirm), but to random failures of opponents, which have nothing to do with the game (missing one or two tournaments).

Therefore, I proposed reducing the aging coefficient to 1/10, which will increase the elasticity of the rating and prevent relatively random events from significantly affecting the ranking table. Stronger players will dominate weaker players for longer. This is what the rating should reflect, in my opinion.

Otherwise, the rating will only reflect the last one or two tournaments, which, if a leader doesn't participate, automatically deprives him of the right to be considered a strong player. But is that really the case? Naturally, refusing to compete gradually negates the player's rating (the rating will halve after six missed tournaments). Which seems closer to reality.

Note: by proposing to reduce the aging coefficient, I did not mean that aging would take place after each ranked tournament across all versions of the game. It was about only those tournaments in which the player can participate. If a player participates in a doubles tournament, then tournaments in a four-player format do not affect his rating. By “each” tournament, I meant that there would be no delays of a month and indulgences for missing a tournament for anyone – after a tournament is missed for any reason (within their version), the player’s rating must be recalculated. That is what was meant.

1.2. General current rating. Combining players into a single rating makes sense only when applying a uniform approach to conducting tournaments in different versions. With a different number of tournaments in different versions – which is the case, it is incorrect to compare players with each other due to the possible positive dynamics for some and "stagnation" for others, and this is not the player's fault. You can reflect the current rating, but it should not be compared; i.e., a player who occupies a higher position relative to his opponent from another version should not be considered a stronger player. Only players from one version will be compared with each other. In this case, the absurdity of bringing all players into a single table becomes obvious at first glance. This will not bring the expected effect from a unified rating.

2. Cyclic rating. It is maintained for a certain period of time, after which results are summed up, a winner is determined, and the result is reset.

2.1. It does not undergo an aging procedure, consists of the sum of tournament ratings for tournaments in which the player participated, accumulates its value over a specific and predetermined period, and is reset after the final results are announced.

3. Tournament rating.

Of the three proposed options, there were positive comments about option #1 (non-linear calculation based on comparing the leader’s result with the player’s result), as well as option #3 – an oscillator designed to adjust the first option depending on the total number of participants in the tournament, as well as their mutual positions. That is, the absolute result of a player does not guarantee him a fixed rating. Everything will also depend on how many participants his result surpassed in the tournament.

For example, completing a map in 25 days with a leader's result of 8 days will give the player a fixed 32 points (in a 100-point scoring system) according to the first option. In the case of an oscillator, if the player took second place at the same time (i.e., there is a large gap from the leader), then the oscillator will correct the result upwards due to the strong performance of the player. And if, at the same time, the player took one of the last places, the oscillator will hardly make any adjustments for this player. In other words, the player's result in the tournament is more objectively assessed, which is not the case in the calculation of option #1. Option #2 is linear and serves auxiliary functions for the oscillator.

I agree that the second (linear) component of the oscillator has an unequal impact on the participants in the tournament table, but this can be adjusted. The question is – should we introduce an oscillator at all?

4. Scale. It was proposed to introduce a 100-point scale for evaluation in order to have a more convenient mechanism for calculating (linking to a percentage scale) and possibly abandoning fractional numbers, i.e., operating only with integers.

5. Minimum guaranteed rating. It was proposed as an option to introduce a minimum guaranteed rating for participation. For example, 10 or 20 points. IronAxe’s position was opposite – it was argued that a bonus for participation could be offered through grimoires or something else, but the game rating has nothing to do with it. I agreed, recognizing this argument as convincing.

P.S. By the way, refusing a minimum guaranteed rating solves the problem with the oscillator. There is no longer a question of "pulling up" an outsider undeservedly high because both parameters are calculated from zero, and their average sum also tends towards zero...

6. Tournament qualification. It was proposed to introduce a coefficient reflecting the qualification of each individual tournament depending on the composition of the participants. To calculate it, the current rating of players will be used and compiled in the form of a quotient obtained by dividing the average rating of a certain number of leaders (for example, 5) participating in the tournament by the average rating of all participants in the tournament. The essence of this coefficient is: the higher the strength of the competitors, the higher the coefficient, the more significant the result of the player shown in the tournament. The coefficient increases the number of rating points awarded to the player for the tournament. This can somehow compensate for the imbalance caused by the appearance of a strong player in the tournament, who significantly “lowers” the bar with his exceptionally strong performance....

Here are the main points of the conversation briefly outlined.
All that remains is to make a final decision on each of the issues:

1. The value of the aging coefficient is 1/10.
2. Rating scale – 100-point.
3. Is it advisable to combine players into a single current rating if there is no tournament parity?
4. Final formula for calculating the tournament rating (non-linear formula or oscillator).
5. Minimum guaranteed rating.
6. A coefficient that increases the qualification of the tournament and its impact on the calculation of the rating.

P.S. And one more thing. Outside the discussion of the rating, there is a proposal, as an exception, not to apply the reducing coefficient to participants in the past BB tournament at the "Normal" level, but to take their result into account in the overall results (without depriving any of the "Impossible" players). It must be admitted that saving on "Normal" was no easier than saving on "Impossible," and sometimes even harder... :) But this issue remains within the competence of the respected VDV_forever. Still, rules are rules...

Well, essentially, almost everything is stated correctly except for one point which I will elaborate on below. So, two independent but mutually influencing ratings are under consideration.

Ok. Everything was presented logically and systematically. But what happens if a monster reaches 1000? You still haven't answered :)
Of course, one could try to argue about the decay by taking an average between 1/3 and 1/10. It will be slightly harder to calculate, but it's essentially the same as 1/3.
And another thing, why do you say that some random player will break ahead if the rating leader misses a couple of Tournaments? The one who was nipping at their heels will move forward. This will happen regardless of the decay value. Seems like no one else is talking. It's not because the system has become so complex that people no longer get it or don't want to get it :)

I haven't heard from anyone else lately. Is it because the system has become so complex that people either can't or don't want to figure it out?

:D :D :D
Or maybe it's because they're afraid we'll ask for Grails as payment for such work – five pages of pointless chatter, that's quite a task! But we won't ask for much...

1. Regarding a rating of 1000. Let someone try to achieve that... I doubt it's even possible in principle; it's a theoretical limit. To reach it, you'd have to play at least 100 tournaments in a row and always win! I thought you were joking when you asked what I would do with such players... There's nothing left to do with such players in Heroes... :)

Firstly, and secondly, if a qualifying tournament coefficient is introduced, then 1000 won't even be the limit anymore...

2. A random player... For example, if all five leaders in the ranking don't participate in a tournament, they receive a -1/3 penalty to their rating, and the former sixth-place player, who just played in a tournament but is just an average player, ends up at the top. That is, it's not his merit that he's in the lead, but the fact that others didn't play... Or something like that... The main thing is that I'm trying to convey the meaning that the player's personal merit should be more significant, but it turns out that the participation or non-participation of a competitor determines everything... This is what I called an imbalance during our conversation.

Well, in any case, we've covered most of the important points from two perspectives. We'll see how things turn out.

:D :D :D
But hey, maybe that's the case... Or maybe they're afraid we'll ask for Grails for such work – five pages of pointless chatter, that takes some effort... But we won't ask for much...

:) There's nothing more to take – they haven't started the printing press yet :)

1. Regarding a rating of 1000. Let someone try to achieve that... I doubt it's even possible, it's a theoretical limit. To reach it, you need to play at least 100 tournaments in a row and always be the winner! I thought you were joking when you asked what I would do with such players... There's nothing to do with such players in Heroes anymore... :)

Don't mislead the grateful listeners :)
10 tournaments * 100 = 1000
They'll reach a thousand in a year and start yelling that they're playing against bots, who came up with the rating, and then they'll bring in Tyapkin and Lyapkin, Lyapkin will hit Tyapkin, and Tyapkin will hit Lyapkin :)

Don't mislead the attentive listeners :)
10 tournaments * 100 = 1000

Why aren't you using the infamous aging mechanic? After all, you need to apply -10% after each round, so recalculate it... What will you get then? You will only score 999 points after completing 66 tournaments... And I won't even mention 1000 :)

I haven't heard from anyone else lately. It's not because the system has become so complex that people can't figure it out or don't want to.

I went to Moscow, and now I'm sitting here reading your discussion. I must say, you've developed the idea well and figured out what everyone is talking about.
Regarding your suggestions, I'll express my opinion.
1. I support aging by 1/10 per tournament.
And here's why. Let's take 2 players. 1 - won the first 2 tournaments and skipped the 3rd, 2 - played in all 3 tournaments, performing decently - 50 points per tournament. What do we have with aging at 1/3: the first has 111 points, and the second has 108 points.
And if the aging is 1/10: the first will have 171 points, and the second? - 135! This is a slightly fairer ratio.
2. Everything is clear here.
3. The oscillator is a rather ambiguous thing that can both add and subtract points. But I'm still for it.
4. I'm for a 100-point scale.
5. A minimum guaranteed rating is not needed.
6. As I understand it, this coefficient will increase the points for all participants in the tournament, including the winner => his point difference from everyone else will become even greater. What is the point of introducing it? How can this coefficient then, to quote: "somehow compensate for the imbalance caused by the appearance of a strong player in the tournament, who will significantly 'lower' the bar with his exceptionally strong performance...." If I misunderstood something about the coefficient, please explain.

6. As I understand it, this coefficient will increase the points for all participants in the tournament, including the winner, which means that his point difference from everyone else will become even greater. What is the point of introducing it then? How can this coefficient, as you say, "somehow compensate for the imbalance caused by the appearance of a strong player in the tournament, who will significantly 'lower the bar' with his exceptionally strong performance?" If I have misunderstood something about the coefficient, please explain.

I think it needs to be clarified.

1. Let's assume we are playing a tournament where all opponents are equal. The winner (22 days) receives 100 points. Second place, 25 days - 22/25*100=88 points.

Another tournament, let's say on the same map (for simplicity) but with different players. And a strong player joins the tournament, who completes it in 8 days and scores 100 points. The others complete it similarly to the players in the previous game: 22 days - 8/22*100=36 points, 25 days - 8/25*100=32 points.
This is understandable.

A small imbalance arises - two different players completed the same map in the same time (22 days), but in the first case, the opponents were easy, and the player managed to get 100 points, while in the second case, a player of the same skill level was given only 36 points (for second place!) due to the appearance of a strong opponent. This is what I call "lowering the bar"... This is somewhat unfair... I say "somewhat" because in life, everything is not always fair - some are luckier than others, which is what we see in our example...

The oscillator is also designed to combat these kinds of inconsistencies - second place should still be significant, even if the gap from first place is very large. We will still have to fight for its adoption, but that is a matter for the future :)

2. So, here's what I'm getting at:
if a tournament qualification coefficient is introduced, then the appearance of a strong player (or several) in the tournament increases the rank of the tournament itself. Moreover, the dependence is direct - the stronger the opponents, the higher the coefficient. That is, in the case of a strong player appearing, for a map completed in 22 days, not 36 points will be awarded, but quite possibly 60 points, or even more...

Which will somehow compensate for the fact that the bar was significantly lowered by a strong player. The winner will also receive a higher rating, but we are comparing two players of equal strength, one of whom was simply "unlucky" with his opponents...

Well, in short, it will look something like this... In sports, by the way, this is exactly what happens. To get the next title, you need to excel in a tournament where there are definitely masters of a high rank, otherwise, a victory among relatively weak opponents will not bring the athlete anything but a medal... :)

Added 20 minutes ago

3. The oscillator is a rather ambiguous thing that can both add and subtract points. But I am still for it.

But it will do so fairly and eliminate the effect of randomness in the outcome, which is very important.

As I understand it, this coefficient depends on the ratings of the tournament participants. I would just like some clarification on this: is this rating the current rating of players from previous tournaments, or only the rating earned by players in this tournament, without taking the coefficient into account?

Added 2 minutes later

But it will do so fairly and eliminate the effect of randomness in the matchups, which is very important.

That's why I decided to support the proposal to use it.

Is this rating the current player rating from previous tournaments, or just the rating earned by players in this tournament without considering a coefficient?

The calculation uses only the current rating (which takes into account all the player's tournaments), which, in theory, should reflect the REAL strength of the participating players. That's why I'm fighting to eliminate all random factors... The rating earned by players in this tournament still has some random character, because one result is not an indicator of strength...

But why don't you use the infamous aging? After all, -10% needs to be applied after each round, so recalculate once more... What will that result in? You will only reach 999 points after completing 66 Tournaments... I won't even mention 1000 :)

Well, I did say I'm not strong in mathematics :) I didn't take into account that the higher the leader's accumulated rating becomes, the smaller portions they will receive from victories over time due to aging, and that is correct.
But still, everyone should agree on aging now; otherwise, you said in PMs that it was right that you were against aging :) Or did you mean 1/3. I maintain my position that 1/3 is more dynamic in my opinion. But I would also agree to 1/10, BUT NOT LESS :)

The calculation uses only the current rating, which, in theory, should reflect the REAL strength of the participating players. That's why I'm fighting to eliminate all random factors... The rating earned by players in this tournament still has some random character because a single result is not an indicator of strength...

Then I'll try to explain why I don't like this coefficient.
1. As I understand it, this coefficient depends on the ratings of the tournament participants. Moreover, its value depends not only on the rating of the leading players but also on the players whose ratings are not in the numerator of the coefficient. Suppose that two tournaments included players (let's call them leaders) whose ratings are in the numerator, and in both cases, the numerators turned out to be almost the same. BUT: in the first tournament, many players with slightly lower ratings than the leaders participated, while in the second tournament, there were many players with low ratings relative to the leaders; as a result, the coefficient in the first case will be less than in the second, even though there were more strong participants in the first tournament. Thus, in the second tournament, everyone will be awarded more points than in the first, despite the weaker composition of participants.
2. A player who "fell below the bar" may have just joined HeroesWorld and have a rating of 0. Then this coefficient will not take such a player into account as you intended.

The calculation uses only the current rating (which takes into account all of the player's tournaments), which, in theory, should reflect the REAL strength of the participating players. That's why I'm fighting to eliminate all random factors... The rating earned by players in this tournament still has some random character, because one result is not an indicator of strength...

But if he stumbles, then at least the one who was right behind him will be able to overtake him. Or he'll have to wait for the leader to die :)

Added 6 minutes later

2. A player who "fell below the bar" may have just joined HeroesWorld and have a rating of 0. Then this coefficient will not take such a player into account as you would like.

No one is immune to surprises. You can't account for everything. But then this player will get a good rating and in subsequent Tournaments will be taken into account as it should be :)

1. As I understand it, this coefficient depends on the ratings of the tournament participants. Moreover, its value depends not only on the rating of the leading players, but also on the players whose ratings are not included in the numerator of the coefficient.

But look at the situation from the perspective of the average player – in which tournament will it be easier for him to take a certain place? In a tournament with more strong players or one with fewer? And so on – for each player. It is more difficult to take ANY place by competing with strong players than to take the same place by competing with weak players... therefore, the tournament rank should be taken into account...

...in the end, we will get a coefficient that is smaller in the first case than in the second, even though there were more strong participants in the first tournament...

The absolute strength of the players is not important, but the relative strength is! Let's say that if only mega-dinosaurs, the strongest players, participate in one tournament, then the tournament coefficient is equal to 1. In the second tournament, only beginners participate. The coefficient will also be close to 1. Because everyone is equal, and no one is competing with a stronger player, and in both cases, everyone has equal chances of winning.

Now let's combine the two groups. Beginners have much less chance, and the higher they get in this situation, the more significant their result will be. Do you understand? So, in reality, we always have a combined scenario – there is always a difference in strength, and it is rare when there are several equal players... Moreover, almost every player has opponents who are much stronger than him... The higher the average strength of the opponents, the more difficult it is to succeed in the tournament... This is what the coefficient reflects – the reward for achieving it :)

This is just an idea. It can and should be adjusted as we go. I'm all for it! The main thing is that the balance is as realistically as possible, close to the golden mean, taking into account all interests. As an option, instead of adding a multiplier, we can add a summand. That is, add a single bonus to the rating that is the same for everyone (and not multiply), but the bonus will depend on the tournament rank. And for the winner, the significance of the bonus in the total of his points will be significantly less than for the underdog – which is also correct. The winner had almost no competition compared to the others... Or make the multiplier non-linear... :) In essence, if you don't go into the details, it seems like it's a lot of complications on top of complications... But in reality, the machine will calculate everything. Our job is to develop a methodology... And I think this will be the right approach, so let's keep it in mind...

2. A player who has "fallen short" may have just joined HeroesWorld and have a rating of 0. Then this coefficient will not take such a player into account as you would like.

At the start of the rating, there will be discrepancies. Over time, when there are enough players with a formed rating, such a player will not have a significant impact on the coefficient in total...

Then this coefficient will not take such a player into account as you would like.

The coefficient will never take all factors into account. Our goal is to reduce the unaccounted moments as much as possible.

The coefficient will never account for all factors. Our goal is to reduce the unaccounted factors as much as possible.

Well, it seems logical. Let's take it into account. Or should we listen to the head of the transport department some more? Where are the administrators and moderators?