It looks like you're new here. If you want to get involved, click one of these buttons!
Tonight on GeekNights, we discuss the high level difficulties of matchmaking in multiplayer games, and why it's so important. We've covered this in panels before, but never done a show on it, and now we're suffering mightily in Rocket League. In other news, Natural Selection 2 is queuing some big changes, Rym extols the virtues of Undertale, and Civ V continues to be the most dangerous "one more turn" game in the world as we play against humans
Download MP3
Comments
Since then I have gotten significantly better, and the main reason is that I actually bothered to play a few multiplayer games. To use a poker analogy, I think the key difference is that I shifted my play style from tight-passive to tight-aggressive.
So yeah, I'm totally up for future games with you peoples.
Rym's Glory to Rome board game analogy was really accurate, he and Scott taught me and another noob at PAX Aus how to play but neither of the 2 noobs were playing it was literally just Scott vs Rym. It wasn't a great experience and I didn't end up learning any intricacy to the game.
The Starcraft 2 formula is kind of broken, I am still ranked as a Diamond player according to Blizzard but I haven't played the game for 4 years. There has to be time decay at high ranked play or the player base needs a soft rank reset every year.
The only downside with ranked Counterstrike is that you can go into a ranked game with a group of people, less than 5. So if you have a 3 assholes who have had their prior Steam accounts banned, they let you win the game for them then kick you before the end of the game. Otherwise it's fine. For Counterstrike the max number of players to go into a 5v5 should be 1 or 2 and if you duo you should be put in a higher difficulty bracket.
For players who want to grind, give them some superficial separate award that is clearly different from their skill ranking.
Everyone hates MOBAs but Riot Games does all these things for their ranked player base. If you are particularly talented, the more ranked games you win in a row the faster your rank increases and then slows down when you start losing games and tries to keep you at a 50% win rate. I'm pretty sure Counterstrike has employed most of these things.
In general though that is frustrating from a matchmaking perspective. The game doesn't do that great of a job of rewarding teamwork which is something that REALLY needs to happen. But there is a fine line between rewarding the player who is top on a losing team and at least acknowledging the players who were not. I think there is like +/- 1 point changes in point loss for top player on a losing team but that's not much. And they already have issues of people wanting to achieve that spot for their team. So you get a bunch of yahoos who all want to me MVP on their team but do dumb shit like crowd the ball into the corner and can't get it out or people who can't practice positioning on the pitch and are never in a position to save a runaway ball.
I think one of the biggest issues with the game honestly is people just don't understand soccer. I remember hearing somewhere the players from one of the top teams talking about playing with randos and them saying that they're often just sitting in the back because they can't trust their teammates. They also talked about how new players are usually really bad with positioning and have a tendency to clump together. It may be partially a matchmaking issue trying to separate out serious people and non serious people. But at the same time there are issues with the playerbase itself who just don't understand how the inspiring game works. Imagine if everyone on a soccer team chased the ball into the corner in the same way. That would be disastrous and idiotic. The same thing happens in Rocket League.
I understand what you're trying to say but the matchmaking should account for those people who understand the basic concepts of what occurs on the pitch.
I didn't experience a matchmaking system in that game.
It felt as if it was just randoms every game.
This does leave the issue of potentially requiring many games to get a signal out of the noise. Thankfully rocket games are short.
Population rapidly dwindled as anyone who didn't have time to play constantly and get top rank(so, most people) left. It was already dying by the time they started fracturing the community with DLC, which was the final blow.
Literally seperates game playlists into Social and Ranked. Rank is based on True skill ranking system (50 levels). If you lose 2 games in a row your level drops 1 , if you win 2 it goes up 1.
The social playlists levels awards 1 exp per won game. Which represents how much of that gametype you've played. But not skill.
Bungie tried something different for Halo Reach, which really wasn't good. They ended up abandonning it early on for something just as stupid. I can't even explain the system. It assumed too many things. Like, how many people were playing the game.
Your rank would reset seasonally, to give you like a historical rank or something. Which was BS, because unlike in Halo 3 where everyone could visibly see what rank you were in skill and be intimidated, your skill rank was gidden in some dark alley where no one could find.
Making it a pointless nymber you couldn't brag about.
The Halo 3 system was dead simple. You win enough, you get a point, you lose enough, you lose a point. There was rarely a chance you would be playing someone out of your league.
Even then, it was a fun challenge.
Rage quitting penalised a whole XP if I remember correctly. That 1 measily xp took a whole game to earn, so you wouldn't quit unless there was some real BS going on. Like teamkills. Then you could report players anyhow.
Those sweet sweet player bans. Satisfaction.
I miss those Halo 3 days.
Consider this hypothetical scenario. You have 10 players, with "true" skill levels of 1, 2,3,4,5,6,7,8,9, and 10 with 10 being the best player. But you don't know their true skills, because they all start at 0.
So you get a game that is 10,7,2 vs. 3,4,5. 10,7,2 wins. Judging only by wins and losses you think that the player with 2 skill is as good as the player with 10, and you can't tell that the player with 5 skill is any better or worse than the player with 3.
You are correct that if these 10 players keep playing a very large number of games against each other in lots and lots of combinations, and you judge only by wins and losses, eventually you will be able to determine their "true" ranks.
But that is because it is a closed system with no new players entering and no players leaving. In a large online game you are constantly facing new opponents. Every single game you play will be determined by the single strongest player, and you will incorrectly adjust the ranks of the other players based on which team that player happened to be on.
Yes, that very very strong player may be able to fly up through the ranks, but since it is a grinding system it will take them a very long time. In every game they play, that is one player who will correctly increase rank with 5 who just get dragged along for the ride.
Because new players are constantly entering the game, at varying skill levels, you will be able to determine that a level 10 player is better than level 2, but you won't have enough data to determine who is 5 and who is 6.
If you do want to stick with wins and losses as the only measure, at the very least you need to not have a grinding system. Someone who wins their very first game should probably immediately be matched up with the best players in the entire game. When they lose you can put them in a game with the median skill-level players and so on, like a quicksort.
Obviously, that has problems. The pro players will not be happy with nubs showing up in their games. While that solution is somewhat good at figuring out player rankings, it's not good at maximizing player experience. You need to get people into games they belong in as quickly and consistently as possible. And to do that, you need you need to judge players based on more than just wins and losses.
Starting out you get slotted somewhere in the middle, with high uncertainty. At the beginning, as you win or lose you shoot up or down dramatically. Once the system gains confidence about your skill level, your uncertainty goes down and you don't move as much. Enables quick convergence on your "true" skill, for whatever that means.
http://research.microsoft.com/en-us/projects/trueskill/
That is NOT how Rocket League works. In Rocket League you start at the bottom and have to grind your way to the top.
The thing is, you can do even better than True Skill if you used advanced metrics to judge players on more than just wins/losses. That's just more difficult to code, and requires more computing resources, and programmers are lazy.
Edit: I forgot another piece! You can assume everyone entering the system is average, because on average, we are.
Neither Rocket League, nor any video game I know of, even tracks (at least not publicly) or uses the most basic statistics in their ranking, let alone advanced metrics. That's ridiculous considering that to track baseball statistics we have to have human beings watching every single player every single game and manually recording all the data. Occasionaly that recorded data is subjective (error or hit?). For a game like Rocket League we could have absolutely perfect objective recordings every single frame of every single game and digitally extract the statistics from that. If we develop more advanced metrics as time goes on, we can go back and reanalyze all the recording and re-rank players accordingly.
Even if you just used the basic statistics you could drastically increase, as TrueSkill puts it, the number of interesting games. Someone loses a bunch of games because of teammates that score own goals? TrueSkill will rank you down because it will correct itself over time. Why bother? If you just tracked even the most basic statistics you could see that this player scored 3+ goals per game, but lost because of their shitty teammates. Rank them up even though they lost.
But that is actually part of the problem for Rocket League specifically. Historically soccer does not track many stats! It's the most unadvanced of all the major sports in this regard. If FIFA, which has a lot more resources than Psyonix, had developed soccer stats over the years, it would have been a lot less work for Rocket League to implement them.
http://grantland.com/the-triangle/the-adolescence-of-soccer-stats/
There is also the problem that there haven't been many sports games where each player is controlled by one individual human being. One person controls an entire team in Madden. Rocket League proves the concept is very viable. But also, it means that there isn't a history of tracking such metrics in video games. If there was a baseball video game where each human controlled a single virtual player, they could simply implement all the metrics that already exist.
The skills required to win a Splatoon match are also very different from traditional FPS, but there was no "I learned this one skill, so now I am a god." You feel yourself slowly getting better at ink placement, stealth tactics, etc., along with traditional learning curve of map knowledge, weapon matchup strengths/weaknesses, etc.
Mario Kart 8 doesn't use as complex of a matchmaking, but it seems very effective in dumping you, very quickly, into a race with equal skilled players.
Has Nintendo come out of nowhere, from basically not supporting online play at all, to offering something few other companies can? I haven't played much Smash online at all, but can anyone weigh in on how the matchmaking works there?