tag:blogger.com,1999:blog-3064325214589649535.post1287475744177048678..comments2024-03-18T08:42:15.276+01:00Comments on Joost's Dev Blog: Why good matchmaking requires enormous player countsJoost van Dongenhttp://www.blogger.com/profile/00569566310604620045noreply@blogger.comBlogger21125tag:blogger.com,1999:blog-3064325214589649535.post-91182822183654315412016-03-03T09:41:44.479+01:002016-03-03T09:41:44.479+01:00I can imagine that would be a fun thing for a cust...I can imagine that would be a fun thing for a custom game or local botmatch, but for public online matches I have little doubt that most of the community would utterly hate that. Just imagine how someone would feel if they lost because they got a handicap or because the enemy got a buff...<br /><br />Also, some players smurf so it's impossible for the game to be 100% sure whether someone is a beginner/kid, or whether he is deliberately leaving matches to keep his rating low and not get a handicap. It would be very abusable.Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-30720700409892810882016-03-03T02:59:10.475+01:002016-03-03T02:59:10.475+01:00I've been playing Awesomenauts with my kids fo...I've been playing Awesomenauts with my kids for years now. We basically stopped playing online as a team.<br /><br />What I don't understand is: why not introduce handicaps? That way strong player can be matched with weak players, and both still play a challenging match. It probably doesn't even take that much... Maybe just increasing the health based on your handicap would go a long way.<br /><br />I mean: you clearly do that with bots... Why not with real people?Anonymoushttps://www.blogger.com/profile/06508106623614342269noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-89533438037041565532016-02-16T04:46:21.057+01:002016-02-16T04:46:21.057+01:00Yeah, I'm by no means experienced with coding ...Yeah, I'm by no means experienced with coding (basically just dabbling in HTML and CSS...), but I definitely find your style clear and engaging.<br /><br />Funny to hear about those other referrals. I guess the lesson there is that some subset of players will complain about matchmaking even when they have no reason to do so.The Gemsbokhttp://thegemsbok.comnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-83057597737513649452016-02-12T09:38:51.528+01:002016-02-12T09:38:51.528+01:00Good to hear it's so clear! :)
I also occasio...Good to hear it's so clear! :)<br /><br />I also occasionally get visitor bumps on this post when someone links it in a Smite or DOTA2 discussion. Which is quite hilarious as these games have gigantic userbases, so most of these problems hardly apply to them...Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-74893961510570708422016-02-12T00:24:02.877+01:002016-02-12T00:24:02.877+01:00I love these development explanation posts, and re...I love these development explanation posts, and read all of them when they get posted to /r/awesomenauts. But this one is particularly good: a clear and easy-to-understand explanation of a topic that is consistently discussed by players. I revisit and re-read this post from time to time, and refer curious players to it liberally. Thanks for this!The Gemsbokhttp://thegemsbok.comnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-84642680204825933152015-08-29T16:13:16.724+02:002015-08-29T16:13:16.724+02:00This is why server browsing is important. If you d...This is why server browsing is important. If you don't have enough players to adequately flood the matchmaker, having access to a server list is better for a myriad of reasons.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-19489433509102348162015-02-05T16:24:54.675+01:002015-02-05T16:24:54.675+01:00Glad to hear there is plans for better matchmaking...Glad to hear there is plans for better matchmaking. It's frustrating to be rank 1 or 2 and be with players that don't understand how to use their team mates to their own advantage while also the teams advantage. (Eg. a Leon that doesn't pull enemies into yuri's mines and continually 1v3's on other side of map and dies but repeats the same thing like 10 times). And then there is the occasional statement "This is my first time using this character" even though it's rank 1 or 2.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-35298861814967601412015-01-31T23:50:53.797+01:002015-01-31T23:50:53.797+01:00It is a really interesting topic, and I look forwa...It is a really interesting topic, and I look forward to the posts!<br /><br />I think being at this point in player counts, where those purely statistical approaches huge games use start to break down, actually makes for a more interesting set of problems and potential solutions!<br /><br />For instance such measurement of player feedback on the matchmaking process itself naturally lends itself to supervised/reinforcement learning. Can machine learning techniques do better at smaller sample sizes?<br /><br />I wonder if there's been much work done on trying to dynamically model player attributes such as emotions (e.g. happiness/frustration)? You could even augment this with sentiment analysis of the in-game text chat.<br /><br />For instance if your model predicts the player is likely getting very frustrated, match them with happy team-mates to try and calm them down! Or if they lost 5 matches in a row, deliberately put them in a match they are almost guaranteed to win.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-54211834824221709702015-01-31T17:41:35.397+01:002015-01-31T17:41:35.397+01:00Ah, yes, that is indeed a very different definitio...Ah, yes, that is indeed a very different definition of "ideal matchmaking". Making the best system possible for a given real situation is a big challenge. We think we can do much better than we are currently doing with Awesomenauts, which is why we are building an entirely new matchmaking system.<br /><br />Still, even the theoretical best matchmaking system will give poor results if the playerbase is not big enough.<br /><br />Various approaches to building a matchmaking system is an interesting topic for a blogpost, I actually plan to write at least two posts about that at some point. The new matchmaking we are building is already our third approach and we have extensively analyses several other, there are lots of interesting things to say about matchmaking algorithms. :)Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-55866477611929331932015-01-31T17:37:49.150+01:002015-01-31T17:37:49.150+01:00I was trying to frame the definition of "idea...I was trying to frame the definition of "ideal matchmaking" something along these lines: consider an oracle machine that somehow always picks in advance the best possible partitioning of the available players into matches that would maximise resulting player enjoyment of the matches.<br /><br />Then you can define your system's performance as relative to this theoretical "ideal matchmaker". Then "ideal matchmaking" is defined in terms of the players you actually have available which seems more useful.<br /><br /><br />As for what I meant by player rating stats, one way you could aim to measure this performance in the real world would be to collect user submitted ratings of the matches afterwards. For instance you could ask them to give the match scores out of 5 for:<br />- Overall enjoyment of the match<br />- How well balanced the match was in terms of skill<br />- How good or bad the "lag" experience was<br />- How well-matched the languages were, or other geographic considerations<br /><br />Then you have some data to work with that hopefully reflects user experience to some extent. Which lets you measure your development progress against these as you try different ideas, and you can at least in principle optimise these different aspects independently.<br /><br /><br />And for your blog readers you can publish the data answering interesting questions like where is the sweet spot in the tradeoff between minimizing experienced "lag" and minimizing skill difference? Just how long is the average player prepared to wait in a matchmaking queue? :)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-25880155440925206062015-01-30T19:34:20.363+01:002015-01-30T19:34:20.363+01:00For the new matchmaking we are building we are ind...For the new matchmaking we are building we are indeed considering to give more difficult groups a longer waiting time to be able to match them better. This might include premades and people in regions with few players at that time. We have not made the final decision yet on how that is going to work.<br /><br />What do you mean by "player rating"? Isn't that just a number that represents skill? Or do you mean something else?Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-52243350563249512932015-01-30T17:15:43.384+01:002015-01-30T17:15:43.384+01:00Points taken! Have you considered giving more diff...Points taken! Have you considered giving more difficult to matchmake players (e.g. geographically remote) a longer wait in the system?<br /><br />But it depends how you define "ideal" matchmaking. I believe it should ultimately be defined in terms of player experience of the resulting matches. Then skill and geography are ultimately just heuristics to get you there.<br /><br />Player ratings of the resulting matches are the data to which I referred that are missing from this picture against which performance ought to be measured.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-59791546595797968302015-01-29T09:38:03.386+01:002015-01-29T09:38:03.386+01:00I only said 'ideal' matchmaking is not pos...I only said 'ideal' matchmaking is not possible and I showed the numbers in the post.<br /><br />There is indeed geographical clustering. For most players this makes it better, for a smaller group this makes it much worse. In the ultimate case everyone would be in the same geographical cluster. Then all the numbers in the post above would be 3 times lower.<br /><br />As for daily fluctuation: our daily peak is currently around 1500 to 2000 simultaneous players. So even when playing in the biggest region at peak hours we would still be far from what I would consider 'ideal' matchmaking.<br /><br />As for players arriving irregularly: the larger the player base the more regular their arrivals will become. The main source or people arriving at the same moment is a previous match finishing, but for 'ideal' matchmaking this is actually quite bad: I would prefer to matchmake people with different players than with the exact same ones over and over again.Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-6511362248201265982015-01-29T05:00:45.058+01:002015-01-29T05:00:45.058+01:00Saying something isn't possible is a very bold...Saying something isn't possible is a very bold statement!<br /><br />It'd be nice to see some actual data backing up these estimates.<br /><br />If you were to partition those 1000 players into sets of good matches you'd surely find significant size variance owing to the clustering of geographical distribution (peak hours?) and skill distribution. Also average players per minute does not seem a very meaningful measure given its variance throughout the day and the clumping / non-independence of player arrivals. Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-47948260641861120252014-11-26T10:22:05.995+01:002014-11-26T10:22:05.995+01:00Doh! Thanks for pointing that out, I have fixed it...Doh! Thanks for pointing that out, I have fixed it in most of the post (cannot fix the image right now but I'll do that later). Funny thing no one mentioned this before even though this post got around 50,000 views.Joost van Dongenhttps://www.blogger.com/profile/00569566310604620045noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-79789244976213766412014-11-25T20:15:56.996+01:002014-11-25T20:15:56.996+01:00"Consecutive" doesn't mean what you ..."Consecutive" doesn't mean what you think it means. I believe the word you're looking for is "concurrent".<br /><br />Otherwise, great read!! :)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-42816909777277502062014-11-25T17:12:50.806+01:002014-11-25T17:12:50.806+01:00Disclaimer: I frequently play awesomenauts but don...Disclaimer: I frequently play awesomenauts but don't keep up with the community discussions, so I'm guessing this topic has been discussed and/or trolled to death, but I'm legitimately curious. Now the question:<br /><br />What were the decisions behind not making the game free to play? I read your post on why free to play inherently can't optimize for the most enjoyable experience, but there are some existing f2p models that (from my point of view) do really well for both players and developers, mainly Dota and League of Legends. (Interestingly, even though Awesomenauts is a paid game, it still can't optimize for the best experience because it has paid skins, as your reply to the Dota comment suggests.)<br /><br />If there's already a discussion around this, I'd appreciate a link. If not, I'd be interested to hear some of the factors that kept/keep you from going f2p, which I think could potentially solve a lot of these matchmaking problems. Is it because you don't make enough off skins to make it a viable option? Is it because you're considering having dedicated servers (one can wish...), and more users would make that harder at this point? Or maybe the Dota/League f2p models only work well with a much larger number of characters and skins to unlock or buy?Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-9704365574008239162014-11-24T22:51:24.057+01:002014-11-24T22:51:24.057+01:00I think you have to be a fellow blog maker for you...I think you have to be a fellow blog maker for you to not be anonymous.Spy A. Rayhttps://www.blogger.com/profile/06264540699924612487noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-70109728162272267792014-11-24T22:50:26.797+01:002014-11-24T22:50:26.797+01:00Or...you could change matchmaking significantly, t...Or...you could change matchmaking significantly, to suit the needs of the servers instead of making a matchmaking system 'good enough' for one server and garbage for the others. Refusing to do this results in requiring massive numbers of players or who knows what which we simply don't have.Spy A. Rayhttps://www.blogger.com/profile/06264540699924612487noreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-6486277618315745002014-11-24T16:32:09.867+01:002014-11-24T16:32:09.867+01:00Hopefully Awesomenauts could become as huge as Dot...Hopefully Awesomenauts could become as huge as Dota 2 - matchmaking times don't take too long when you have millions of players.Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-3064325214589649535.post-76371888326666415992014-11-23T11:51:34.768+01:002014-11-23T11:51:34.768+01:00yay for comments getting absorbed into the void if...yay for comments getting absorbed into the void if you try to type something with google acount!<br /><br />mmr should give the person playing more freedom. They should have the choice to be able to be matched against higher skilled players, Though it should not disregard the people in his own MMR. Also p2p servers are almost always worse than dedicated ones a eu to us dedicated server you will have a steady 120 ping. where as if it is p2p it will most likely have a steady 300 ping or worse.Anonymousnoreply@blogger.com