LOL Matching Algorithm

 

This is a post written by riot's design ctor Tom Cadwell specifically for Chinese gamers to explain how the matching system works.

At the same time, in order to make everyone better understand the matching system, if you think that you have encountered a particularly unfair match, please reply to the game start time and End Time, we will investigate how the match was completed by the Bureau, and why did the poor players add? To this game.

Many people complained that they could not understand it. I 'd like to make a simplified analogy:

There is a basketball league with several players and a league of about 20 levels.

All players are members of the 10-level league. They are free to match each other, win, upgrade to the 11-level league, and lose to the 9-level league.

Then, each level League starts again, and some people upgrade and some others downgrade. Finally, all the 20-level league matches have the player shujia.

A large amount of data proves that a player's level will make it stable between about three league matches, that is, Kobe is playing the top 20 league matches, in addition, when he and four players in the 17-level League form a team, they will not lose to those in the 17-level league. In addition, if Kobe falls to the 10th level league, he will easily return to the 10th level in 20 games.

Ideally, players are playing with players who have the same experience as themselves. A medium-level player won't match Kobe at all, and Kobe won't match a player who just played the game.

In fact, the classification of the matching system will be more complex and intelligent than this. The competition uses the elo system used by chess.

Add again? FAQ:

Q: Will the system give me some food teammates after I win the game to keep the winning rate at 50%?

A: The purpose of the system is not to maintain your winning rate, but to allow gamers with almost the same level to play together. When you play with players with almost the same level, the winning rate is close to 50%, so the system will not help you.

Q: I only win 100. Why do the system always match players with 600 wins?

A: A winner cannot reflect a person's level. Let's assume that we compare the matching system to a running system. The person who has been practicing for three years is the same as the person who ran for 11 seconds for the first time. The matching system is based on the level rather than experience.

Q: Why do I win 60% of my teammates and 40% of my competitors?

A: The winning rate cannot reflect the level. Matching the system depends not only on whether you win, but also on who you win. As in fact, the winning rate of sky in war3 is not high, but the average winning rate of abuse players is 100%. Players of the same level will win 40% to because of random matching to the opponent ~ 60%.

Q: Why do I think they are doing this?

A:Matching systems provide fair opportunities, not necessarily your desired results.We can pursue system justice, but cannot pre-empt the performance of gamers in a single game.

The system 100% matches Manchester United against Real Madrid, but it cannot be guaranteed that Manchester United won't roll over Real Madrid at. In this case, Cristiano Ronaldo's YTS performance is all dragging its feet. Or Manchester United may win Real Madrid three times in a row. However, the system will only match Manchester United with Real Madrid rather than Manchester United Against Super League Shenzhen. Details about whether Real Madrid wins or Manchester United wins in a certain game depend on the arrangement of troops in that game, on-site play, and tactical intent.

Assuming that the pitfall player is not in your level, he will keep losing his teammates and keep reducing his level. This will let him leave your matching range, so that he can no longer match you. Based on our data, the player's Elo is basically stable in a small range. This is the gap between the Shenzhen team and Real Madrid. It is also the reason why the Chinese national team won the French team and never won the World Cup.

The system cannot provide you with perfect teammates. Players may not be able to play well for many reasons: they cannot use any hero, they do not want to play positions, their gameplay style is not enough to match with their teammates, and they may not wish to hook up in the early stage. However, the probability of such a situation between you and the other players is the same. The system will not favor either party. So if you want to be a perfect teammate, please team up with friends, but you will also meet more powerful opponents.

If you have a good idea about how to determine the player level, but if you want to complain about the player's internal performance to prove that the matching system is unfair, you are in a fight with the windmill. Everyone has different opinions. The system assumes that when you meet the same teammates and opponents, the odds are almost the same. This is what we can do best now.

The following is the body of the article.

Overview:

The purpose of matching the system is as follows:

1. Protect new players from being abused by experienced players, so that there are no new players in the master game.

2. Create competitive and fair game matches to maximize the game pleasure of players.

3. You don't have to wait too long to find your opponent to enter the game.

The matching system tries its best to match players with a level close to that of the players from who they won and their opponents before that. When you defeat your opponent, the system will think you are better. When you lose to your opponent, the system will think you are weaker. Although this is not so fair for a certain game, in the long run, it is quite fair for multiple games: Because good players will always have a positive and positive impact on the game results. We used this method to try it out: a new account is provided for players with high levels of water supply, and then the results are displayed after the game number board. We have proved our ideas through a lot of tests.

In addition, the matching system has some advantages for players who know to team up in advance. If you team up in advance, you will be given more players. We use some clever mathematical methods to solve the unfair matching problem between players in advance and solo players. I even asked two math masters to verify it. They all said to me!

How is the matching completed?

First, the system puts you in the appropriate matching pool-based on the Game Model (matching mode, ranking solo/double, ranking 5, other models, etc)

Then, the system will try to group the Matched persons in the pool to a finer match pool-5 members and 5 members, and low-level beginners and other low-level beginners.

When you are in the matching pool, the system will start to try to find a suitable match, with the goal of matching a game with 50% winning chances for both parties.

Step 2: Determine your strength:

* If you are solo, you can directly use your personal matching score (that is, the ELO value. The matching mode and qualifying match score are different)

* Assume that you team up in advance, and your matching score is the average score of your team, which will be slightly increased based on the size of your team, in this way, you can ensure that you match a stronger opponent to offset your team's advantages. I worked with a PhD in computer Biology (Computational Biology Ph. D) to explore hundreds of thousands of game results and figure out how much advantage a team had in advance. We also made some other adjustments behind the scenes, such as new players and high players, for example, which of the blue and purple players in a certain figure is more advantageous, and so on.

Step 2: Determine your target competitor:

* First, the system will match players that are close to you based on your ELO value. Finally, the system will relax the matching conditions and give you some less perfect matches, because you certainly do not want to never match people.

* New users will get some special protection. Generally, new users will only match other new users (in mature servers, this proportion has reached 99% +. Unless this newbie and a senior player friend team up in advance)

Step 2: confirm the matching:

* Finally, the system matches 10 players of the same level and level to facilitate a game.

* The system will try to balance the team and try to make both sides have a 50% chance of winning. In most of the time, the error will be within 3%-similar to 50/50, 49/51, 48/52. In fact, there is a little difference in the chance to win (Q & A will answer this question), but our research shows that in most cases, this is actually a very accurate pre-commit.

In the long term, how does my matching score (ELO value) be merged?

We used a modified elo system. The basic points of the elo system are to use math to compare two people's points to pre-determine the results of the two players-similar to "A and B games, a WINS 75% of the game ".

Then, the results of the competition came out. If you win, you will receive extra points. If you lose, you will be deducted points. If you win unexpectedly (the system thinks you are more likely to lose), you will win many other scores. Additionally, if you are a new player, you will score more quickly, so that you can quickly enter your level. In the long run, this means that good players will get a high matching score, because they always exceed the system's expectation, they will continue to add points until the system can correctly pre-determine their winning rate.

We changed this system for the team to use in the competition. The basic concept is to get a team ELO value based on all the players in the team. If your team wins, the system will if all the players in the team are "more powerful" than the system pushes and add points. Although there are some problems, they are effective on the whole, especially when players team up in advance.

For example, I have 2000 normal matching mode ELO on servers in North America. Suppose I create a trumpet, even if there is no talent or Rune, there will already be 1800elo when I hit Level 8. This system is not perfect, but it can indeed make players quickly close to their own level.

When you start to play, we also make some minor adjustments to Elo, so that you can enter your level more quickly.

* We have a large number of priority methods to identify a player. Compared with a standard new player, it is more skillful and vigorous. Suppose we find that yes, we will improve his ELO level behind the scenes.

* We will also tell the truth about cainiao beginners.

* The upgrade level will greatly increase your ELO value. This will also help the system differentiate 30-level full-level summoners from low-level summoners.

Suppose you want to know the theory of the elo system and many other details, you can look at this:

Http://en.wikipedia.org/wiki/Elo_rating_system

Http://zh.wikipedia.org/wiki/ELO

Er, wait. How do you deal with team players vs solo (single-row) players?

In most cases, we will match a team of five to another team of five to avoid this situation (almost all ).

We have done a lot of research on "partially" teams and found that the advantages are not as big as we think, so we will also mix them into solo players. We found that a large number of factors will affect the strengths of the team: from the size of the team in advance (for example, 2, 3, 4, and 5) to the level of the team players, to the combination of high-performance cainiao players, to the different levels of players, resulting in different situations, and other subtle factors that must be taken into account. This is more reliable than some point-to-point algorithms we have seen-Mixing Random statistical data into boosting scores-

When we discover these advantages, we know how much ELO value should be increased for the teams that are created in advance to achieve a fair match and determine an appropriate and reasonable mathematical adjustment. The results are surprising in some cases (statistical data is corrected at the same time ).

Although we will not give accurate values, as this is a trade secret, we can tell you:

* A team of five members is only a little better than five passers-.

* Some teams are only slightly better than five passers-.

* Five cainiao teams won't bring too much advantage, but high-performance teams will have a great advantage.

* Teams with high team strength and variance will be stronger, for example, teams with low differences. (In simple terms, the variance reflects the size difference of each element under the same average value. A large variance indicates a large difference, and a team with a high variance is similar to a high-performance team with a low-performance band, the team members with low variance are close to each other .)

* This shows that the carry effect of high-level players (can be understood as leading or thigh) is more powerful than that of low-level players (feeder.

Okay... So why should we match players in advance with non-team players?

There are some reasons:

* This will help the system quickly find the matching score that suits you, so that the system can give you a fair match faster. The working principle is that if you team up, it will reduce the components of luck. If you are in a single row, the quality of your teammates will have a greater impact on your winning or losing. If you team up in advance, you will form a team of players that are almost the same level as you, and you will have a lower chance of randomly encountering male/hookan teammates. Since the results of the game are mostly from the performance of friends with similar levels, rather than random factors, your matching club can reach the exact value faster.

* We hope players can play with their friends, because this will make them more fun. Nor can you set a separate 2-person matching pool or 3-person matching pool for a 5v5 game. You need to combine them to make the system work. We chose to team up five members, because it was fun. If we have a large enough matching pool in the future, we may differentiate five teams from some teams. However, the data tells us that this will basically not increase the fairness of matching, the results are basically the same.

Other common problems:

Q:Why not? Some other details, such as the number of kill to determine my matching score?

A:Because this is biased, and it is difficult to rate the number of kill attacks, when you use a gank hero (like a mouse and an easy master), how many people can kill? In addition, this will make good auxiliary players suffer, because they do not aim to take the head, or even block themselves for their own carry. Finally, players will spend a long time playing the game in order to refresh the data, and then take a large number of farm opponent's heads instead of winning the game. We try our best to put the gamer base level and incentive mechanisms above our efforts to win. We have avoided unnecessary peripheral behaviors that are neither fun nor disturbing the matching system.

Q:I am furious because the matching system always gave me feeders ). Why can't this happen?

A:We did try to prevent this from happening, but suppose you are matched to a very weak player, this also indicates that the matching system matches one or more powerful players at the same time. Based on our research, we found that carry (thigh) plays a more leading role in the team than feeder (giving people a head, pitfall. The reason is that in lol, the benefits of repeatedly killing the same player will decrease, not like other similar games. Our analysis shows that when the average ELO is the same, the ELO value of a player in this team is increased or decreased by 100 (other players decrease/increase to keep the same average score ), the overall team's strength will increase the ELO value by about seven points. This also shows that the role of carry in LOL is more powerful than that of feeder. Indeed, sometimes you will lose this game because of matching feeder, but it is because your team's carry is not powerful enough.

Q:In this case, if I win several disks in a row, will I be matched with some completely insurmountable opponents?

A:Not all. Winning streak leads to an increase in your matching club, and you will constantly encounter stronger opponents. However, we are not helpful in keeping your winning rate at 50%, our goal is only to ensure that the system can correctly pre-Prepare the game results. Finally, you will reach your limit, and you will maintain a general 50% winning rate. Players with higher average scores tend to have a slightly higher winning rate than 50%, because there are many other players who are weaker than them and fewer players than them. Therefore, when matching, it is often a little "downward match ". Top-ranking high-end players often have a 90% winning rate.

Q:How do you design a fixed team? A wow-like arena team?

A:This is a good idea and gives us the opportunity to design a better matching system. We will do this sooner or later and use the new method we developed. We need to test and understand how powerful you are in general (such as your personal points) and agree to create/disband a team at the same time. This is a big project, but we are very passionate about it ~

Q:Assuming that the matching system is so fair, why do I always face a one-sided competition?

A:There are two reasons. First, sometimes the "snowball effect" of LOL is very obvious. Poor early stage performance can make the game feel quite different. Especially for some teams, if they start very well, they will always be very good. We met the same team, winning in the first round of 2nd-5, and losing in the second round with a similar score. The second reason is that players are not playing well, and the team selection is not good. To perform a round-robin match, you need to balance the player level and select a balanced lineup. Sometimes a player chooses a lineup that is better than the scum, for example, five melee DPS, three tanks, two magicians, or no wild hero. In this case, although your team is also very good, the situation is often unbearable.

Q:Why do I sometimes match low-level players as a high-level player? They all seem to be here to give people their heads.

A:This is a headache when a high-level player and a low-level player team up. We hope players can play with their friends, and this is a pleasant experience. However, we do not want to establish the happiness of some people on the basis of the pain of some people. Therefore, we often rate this combination to a higher level to protect new players from abuse by High-level players. Unfortunately, no matter what we do, matching this combination to any game may lead to an unpleasant experience. Therefore, we plan to implement an "unbalanced team-up" queue, which is similar to matching five teams to five.

Q:I got 20 levels, and I was matched with some 10 levels and some 29 levels. What happened?

A:When players of different levels form a team, we use their average level as the matching benchmark. The level does not match the dominant number of players in the system-the matching system usually uses strength to match-but we will try to match players with similar levels. We cannot select players in advance, so we do our best to use the average level. In this computing system, we will regard players of level 30 as level 36, So we usually make a medium-grade player's game not of level 30 players, but sometimes, Level 29 players can be inserted.