DUST Search - Hotfix - Episode Charlie: Scotty must die

Beren Hurin wrote:

CCP Rattati wrote:

Thanks for the feedback.

Traditionally matchmaking is referred to as the "selection of the 32 players in the battle".

For each player, an index (ELO like) is calculated, that is based on the individuals Win/Loss ratio. At the end of each battle, that index is updated by the battles result. A players index will go higher (or lower) until theoretically converging to the index value that where they win exactly 50% of their battles as they are consistently fighting players who are at equal skill.

By selecting 32 with similar indices, theory says that if that selection is robust and effective, the 32 should be similar enough to make a fair fight. However, due to many factors, this selection is very inefficient and results in very disparate players skills as most of you know.

That wouldn't be a problem as long as the teams are balanced, from any team sport it is known that is all right if the good players are properly mixed, even better than just fighting people of your own skill so you can learn from your betters. I play soccer and hate nothing more than one-sided matches with unfair teams, as they are bad for everyone, it's absolutely not fun to win either. That is not the case either with poor Scotty.

Therefore. and I have mentioned that recently, we are working on getting improvements on the teambuilding algorithm instead, basing it on the most effective way on splitting the 32 player pool into as perfectly equal teams as we can at the beginning of the battle.

The current design is to collect around 40 players, of the most similar skill we can without making the wait period too long. If Scotty is working, then they will be similar, if not they will be disparate.

Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed.

The aggregation will be sum so large powerful squads will always be fighting each other. Because of the various number of squads and squadmembers per squad, we may need to use lower ranked units, so as to not exceed 16 on each team. Then the remaining 8 players will be reseeded back into the waiting pool to be matchmade again.

Other known battle issues such as not joining battles that are already over/decided, joining empty battles on one side, "squad free" modes(squads fight squads only, solo vs solo) are also on the plan, but lower priority than team balancing improvements.

Thanks for reading

Have you looked into the Ajusted Plus Minus model for matchmaking rather than an ELO system?

"As well as stupid, Rattati is incredibly slow and accident-prone, and cannot even swim"

Pages: 1 :: [one page]

Author	Thread Statistics \| Show CCP posts - 2 post(s)
Kevall Longstride DUST University Ivy League 1449	Posted - 2014.07.07 18:02:00 - [1] - Quote Hi Rattati. As you'll likely be beginning to ramp up the Charlie requests soon, (no not that kind of Charlie), I think its time that we talk seriously about Scotty the AI. Short version: He couldn't outwit a ZX80 Matchmaking is one of the longest outstanding issues in the game and one that would not only benefit new players if fixed but all players. But obviously before giving you feedback we'd need to know how much of the matchmaking is server side and how much is client. As I understand the history, Scotty's algorithm was designed around a concurrent player base total that we long since stopped seeing. Not enough players = Scotty not working This problem would be bad enough but the fact that the spawn mechanic, especially in Ambush and Domination, is incontrovertibly the product of someone's cat walking over the keyboard at a crucial time, makes the frustration for players jump to levels that only ketamine could level off. We know that in CCP Z's progression system for Legion, meta levels of equipment will be taken into account. But this is of course Legion, with a matchmaking system likely to redesigned from the ground up. So in Dust 514, what criteria can't altered and changed from server side and be used to develop a replacement for Scotty? I for one wouldn't mind a slightly longer wait to get into a game, if that games participants are at my level. The requiremnts for a better matchmaking system in the pub matches for me would need to be: A dynamic system that allows for short dips and peaks of form while allowing longer term changes A system that allows for squads and the varied levels for each member of said squad, as well as the performance of that squad while it's been formed. The total amount of SP a player has in relationship to the average of each player on either side. Finally, take into account the War Points a player has earned as that is the key indication as to the quality of a player and not the KDR. But whatever you do, Scotty must die. CPM1 Candidate CEO of DUST University
John Demonsbane Unorganized Ninja Infantry Tactics League of Infamy 3634	Posted - 2014.07.08 04:03:00 - [2] - Quote I refuse to believe that Scotty has some great core programming that is only thwarted by low player counts. I'm sorry, but he's just plain stupid. No reasonably-well programmed algorithm in existence, no matter how many players it was designed for, would so consistently place 2 squads on a single side and none on the other. It's mathematically impossible. (The godfather of tactical logistics)
Appia Vibbia Molon Labe. General Tso's Alliance 3179	Posted - 2014.07.08 05:23:00 - [3] - Quote You put the word "Episode" in the title. Now I think of Hotfix Charlie having Half-Life's fate Appia Vibbia for CPM1 Empress of Alts
Harpyja Legio DXIV 2020	Posted - 2014.07.09 05:18:00 - [4] - Quote I honestly believe that KDR should be a factor in matchmaking; that and average WP/battle or /death. Why KDR? Because it is a fair statistic. Someone who runs proto 24/7 but has a KDR of 1.5 isn't a stomper; they are just like the majority of players. Someone in a standard suit with a similar KDR has a much fairer battle than when up against a proto stomper (who has a KDR that is much higher). KDR is a statistic on survivability. A high KDR means high survivability and is difficult to kill, whereas a low KDR shows low survivability and is easy to kill. Average KDR of both teams need to be as equal as possible to give everyone a fair battle. Also, WP/battle or /death shows activity. Someone with a high ratio is much more active than someone with a lower ratio. This number should also be a team average where both teams are as equal as possible. This should provide equal teams that participate equally, instead of having one that AFKs more than the other. These two statistics should, in my opinion, provide battles that aren't one-sided. Both teams should have fair gameplay that isn't one sided in terms of stomping and also ensures that both teams are as active and participating as much as the other. "By His light, and His will"- The Scriptures, 12:32 -- "Scouts should fart repeatedly while cloaked"- TechMechMeds
XxGhazbaranxX Eternal Beings Proficiency V. 1669	Posted - 2014.07.09 06:56:00 - [5] - Quote In all honesty I would seriously suggest giving up on matchmaking; it would be too much of a hassle to develop an effective algorithm. I would simply just extend squad sizes to 16 players and be done with it. Include team match option that cycles between Dom/skirm/ambush/ambush OMS and that's that. Deny entry to pubs to any team that has more than 6 people. That should fix a whole lot of problems within the matchmaking system since people with enough experience will stay out of PUBS and prefer team matches and wont be able to stomp with 16 man squads in pubs because they will be denied entry. Plasma Cannon Advocate Dust 514 Survivor
Kevall Longstride DUST University Ivy League 1461	Posted - 2014.07.09 08:35:00 - [6] - Quote The problem with KDR is that it is more often than not a reflection of poor balance mechanic. Look at the problem with tanks. The 50-1 KDR that one tank guy had in while I was running a Candidate/Q&A squad last night, for example. How much of that performance was due to his skill or the fact that a triple repped maddy is near invincible, because the module balance in HAV's is so out of wack at the moment. Should we ever get the vehicles/infantry balance right then and only then should KDR play any part in match matching determination. The above hyperthetical is purely wish fulfilment I suspect. CPM1 Candidate CEO of DUST University
Harpyja Legio DXIV 2021	Posted - 2014.07.09 14:03:00 - [7] - Quote Kevall Longstride wrote: The problem with KDR is that it is more often than not a reflection of poor balance mechanic. Look at the problem with tanks. The 50-1 KDR that one tank guy had in while I was running a Candidate/Q&A squad last night, for example. How much of that performance was due to his skill or the fact that a triple repped maddy is near invincible, because the module balance in HAV's is so out of wack at the moment. Should we ever get the vehicles/infantry balance right then and only then should KDR play any part in match matching determination. The above hyperthetical is purely wish fulfilment I suspect. I don't see that as much of a problem. Someone who gets artificially high KDRs not through their own skill but from abusing game mechanics will get put into battles facing similar opponents that may actually have skill. It'll be like having a booster matched up against a proto stomper; the booster will simply stand no chance. If people realize that having such artificial KDRs will ultimately bring them down, there won't be a point to KDR boosting in the first place. Edit: also reps were nerfed. 2 complex and 1 enhanced (or was it the other way around) now provide slightly more reps than three basic pre-nerf. "By His light, and His will"- The Scriptures, 12:32 -- "Scouts should fart repeatedly while cloaked"- TechMechMeds

CCP Rattati C C P C C P Alliance 3679	Posted - 2014.07.09 19:03:00 - [8] - Quote Thanks for the feedback. Traditionally matchmaking is referred to as the "selection of the 32 players in the battle". For each player, an index (ELO like) is calculated, that is based on the individuals Win/Loss ratio. At the end of each battle, that index is updated by the battles result. A players index will go higher (or lower) until theoretically converging to the index value that where they win exactly 50% of their battles as they are consistently fighting players who are at equal skill. By selecting 32 with similar indices, theory says that if that selection is robust and effective, the 32 should be similar enough to make a fair fight. However, due to many factors, this selection is very inefficient and results in very disparate players skills as most of you know. That wouldn't be a problem as long as the teams are balanced, from any team sport it is known that is all right if the good players are properly mixed, even better than just fighting people of your own skill so you can learn from your betters. I play soccer and hate nothing more than one-sided matches with unfair teams, as they are bad for everyone, it's absolutely not fun to win either. That is not the case either with poor Scotty. Therefore. and I have mentioned that recently, we are working on getting improvements on the teambuilding algorithm instead, basing it on the most effective way on splitting the 32 player pool into as perfectly equal teams as we can at the beginning of the battle. The current design is to collect around 40 players, of the most similar skill we can without making the wait period too long. If Scotty is working, then they will be similar, if not they will be disparate. Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed. The aggregation will be sum so large powerful squads will always be fighting each other. Because of the various number of squads and squadmembers per squad, we may need to use lower ranked units, so as to not exceed 16 on each team. Then the remaining 8 players will be reseeded back into the waiting pool to be matchmade again. Other known battle issues such as not joining battles that are already over/decided, joining empty battles on one side, "squad free" modes(squads fight squads only, solo vs solo) are also on the plan, but lower priority than team balancing improvements. Thanks for reading "As well as stupid, Rattati is incredibly slow and accident-prone, and cannot even swim"

Harpyja Legio DXIV 2024	Posted - 2014.07.09 19:06:00 - [9] - Quote CCP Rattati wrote: Thanks for the feedback. Traditionally matchmaking is referred to as the "selection of the 32 players in the battle". For each player, an index (ELO like) is calculated, that is based on the individuals Win/Loss ratio. At the end of each battle, that index is updated by the battles result. A players index will go higher (or lower) until theoretically converging to the index value that where they win exactly 50% of their battles as they are consistently fighting players who are at equal skill. By selecting 32 with similar indices, theory says that if that selection is robust and effective, the 32 should be similar enough to make a fair fight. However, due to many factors, this selection is very inefficient and results in very disparate players skills as most of you know. That wouldn't be a problem as long as the teams are balanced, from any team sport it is known that is all right if the good players are properly mixed, even better than just fighting people of your own skill so you can learn from your betters. I play soccer and hate nothing more than one-sided matches with unfair teams, as they are bad for everyone, it's absolutely not fun to win either. That is not the case either with poor Scotty. Therefore. and I have mentioned that recently, we are working on getting improvements on the teambuilding algorithm instead, basing it on the most effective way on splitting the 32 player pool into as perfectly equal teams as we can at the beginning of the battle. The current design is to collect around 40 players, of the most similar skill we can without making the wait period too long. If Scotty is working, then they will be similar, if not they will be disparate. Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed. The aggregation will be sum so large powerful squads will always be fighting each other. Because of the various number of squads and squadmembers per squad, we may need to use lower ranked units, so as to not exceed 16 on each team. Then the remaining 8 players will be reseeded back into the waiting pool to be matchmade again. Other known battle issues such as not joining battles that are already over/decided, joining empty battles on one side, "squad free" modes(squads fight squads only, solo vs solo) are also on the plan, but lower priority than team balancing improvements. Thanks for reading Thanks for posting, cheers "By His light, and His will"- The Scriptures, 12:32 -- "Scouts should fart repeatedly while cloaked"- TechMechMeds
Beren Hurin Onslaught Inc RISE of LEGION 2447	Posted - 2014.07.09 19:17:00 - [10] - Quote CCP Rattati wrote: Thanks for the feedback. Traditionally matchmaking is referred to as the "selection of the 32 players in the battle". For each player, an index (ELO like) is calculated, that is based on the individuals Win/Loss ratio. At the end of each battle, that index is updated by the battles result. A players index will go higher (or lower) until theoretically converging to the index value that where they win exactly 50% of their battles as they are consistently fighting players who are at equal skill. By selecting 32 with similar indices, theory says that if that selection is robust and effective, the 32 should be similar enough to make a fair fight. However, due to many factors, this selection is very inefficient and results in very disparate players skills as most of you know. That wouldn't be a problem as long as the teams are balanced, from any team sport it is known that is all right if the good players are properly mixed, even better than just fighting people of your own skill so you can learn from your betters. I play soccer and hate nothing more than one-sided matches with unfair teams, as they are bad for everyone, it's absolutely not fun to win either. That is not the case either with poor Scotty. Therefore. and I have mentioned that recently, we are working on getting improvements on the teambuilding algorithm instead, basing it on the most effective way on splitting the 32 player pool into as perfectly equal teams as we can at the beginning of the battle. The current design is to collect around 40 players, of the most similar skill we can without making the wait period too long. If Scotty is working, then they will be similar, if not they will be disparate. Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed. The aggregation will be sum so large powerful squads will always be fighting each other. Because of the various number of squads and squadmembers per squad, we may need to use lower ranked units, so as to not exceed 16 on each team. Then the remaining 8 players will be reseeded back into the waiting pool to be matchmade again. Other known battle issues such as not joining battles that are already over/decided, joining empty battles on one side, "squad free" modes(squads fight squads only, solo vs solo) are also on the plan, but lower priority than team balancing improvements. Thanks for reading Have you looked into the Ajusted Plus Minus model for matchmaking rather than an ELO system?

Grimmiers 622	Posted - 2014.07.09 19:47:00 - [11] - Quote @rattati Sounds good. I would much rather have a longer wait time for a balanced match than the almost instant one we have now that throws me into a losing match about to end. Make it so if Shields are down, or clones are less than 100 the match will be less likely to let someone join.

CCP Rattati C C P C C P Alliance 3710	Posted - 2014.07.09 20:41:00 - [12] - Quote Beren Hurin wrote: CCP Rattati wrote: Thanks for the feedback. Traditionally matchmaking is referred to as the "selection of the 32 players in the battle". For each player, an index (ELO like) is calculated, that is based on the individuals Win/Loss ratio. At the end of each battle, that index is updated by the battles result. A players index will go higher (or lower) until theoretically converging to the index value that where they win exactly 50% of their battles as they are consistently fighting players who are at equal skill. By selecting 32 with similar indices, theory says that if that selection is robust and effective, the 32 should be similar enough to make a fair fight. However, due to many factors, this selection is very inefficient and results in very disparate players skills as most of you know. That wouldn't be a problem as long as the teams are balanced, from any team sport it is known that is all right if the good players are properly mixed, even better than just fighting people of your own skill so you can learn from your betters. I play soccer and hate nothing more than one-sided matches with unfair teams, as they are bad for everyone, it's absolutely not fun to win either. That is not the case either with poor Scotty. Therefore. and I have mentioned that recently, we are working on getting improvements on the teambuilding algorithm instead, basing it on the most effective way on splitting the 32 player pool into as perfectly equal teams as we can at the beginning of the battle. The current design is to collect around 40 players, of the most similar skill we can without making the wait period too long. If Scotty is working, then they will be similar, if not they will be disparate. Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed. The aggregation will be sum so large powerful squads will always be fighting each other. Because of the various number of squads and squadmembers per squad, we may need to use lower ranked units, so as to not exceed 16 on each team. Then the remaining 8 players will be reseeded back into the waiting pool to be matchmade again. Other known battle issues such as not joining battles that are already over/decided, joining empty battles on one side, "squad free" modes(squads fight squads only, solo vs solo) are also on the plan, but lower priority than team balancing improvements. Thanks for reading Have you looked into the Ajusted Plus Minus model for matchmaking rather than an ELO system? No I have not , but our index works very well to predict quality. Our top ranked players by index are very strongly correlated to Warpoints and KDR. "As well as stupid, Rattati is incredibly slow and accident-prone, and cannot even swim"

Kevall Longstride DUST University Ivy League 1465	Posted - 2014.07.09 20:48:00 - [13] - Quote Good to know and thanks for the reply. I understand how Scotty is supposed to work using the method you described but why is he not working as envisioned? What element is causing the problems that require a fix? Just curious. CPM1 Candidate CEO of DUST University
Jadd Hatchen Kinda New here 584	Posted - 2014.07.09 20:49:00 - [14] - Quote CCP Rattati wrote: Then all squads and units will be aggregated into team units. The parameter aggregated will be Lifetime Warpoints/Lifetime Battle seconds, Lifetime Warpoints/Lifetime Deaths or Lifetime Kills/Lifetime Deaths. The units are then ranked and distributed. Hmm... If you assume a specific "normal" behavior for every player to be the same goal, then this sort of algorythm works. That goal being to win the match. However that is NOT in fact every player's goal! Take the simple example of the sniper who's only goal is to maintain a high KDR. Now two of the factors you are using become very heavily skewed as they end up dying much less than everyone else on the team, but overall contribute nothing to the overall goal of winning the match as they never capture or take any of the objectives. Only one of the three factors you have above (Lifetime Warpoints/Lifetime Battle seconds) is low in this case while the other two become skewed very high. This creates a situation where the one player "seems" to be very valuable for matchmaking, but actually sets that team up to loose the match instead. This same mechanic is skewed even worse by HAV pilots (and to a smaller extent by Sentinel Dropsuits). The HAV runs around the game accruing many kills per death and many warpoints per death. But unlike the sniper situation, they ALSO accrue many warpoints during the match for destruction of enemy turrets, vehicles, and supply depots. but all the while they may never help to capture or defend any of the objectives. They are again just looking to advance KDR and Warpoints only. Who wins the battle doesn't matter to them. This again results in the creation of a team that is doomed to never win a match. The characteristics you are using to do matchmaking do not take into account the psychology of your players! This is the heart of why your matchmaking system has always failed. So how do we fix this? You need to add keep track of the Win/Loss record as a fourth factor in your ELO system. Perhaps downgrade the KDR portion of it as well somewhat.
XxGhazbaranxX Eternal Beings Proficiency V. 1669	Posted - 2014.07.09 23:51:00 - [15] - Quote VIDEO Hello guys. Here I leave a small video of how the current matchmaking system is working to enhance the conversation about it's change. As rattati said, they are working on fixing matchmaking and the video is just to give an example of what matchmaking tends to do which is squad stack. Plasma Cannon Advocate Dust 514 Survivor
Garth Mandra The Southern Legion Final Resolution. 412	Posted - 2014.07.10 01:00:00 - [16] - Quote Sounds like progress. By the way it's Elo not E.L.O.. Arpad Elo is the guy that came up with the system for chess.
THUNDERGROOVE Fatal Absolution 997	Posted - 2014.07.10 01:09:00 - [17] - Quote What you're asking for would severely increase battle finding times. I wouldn't really mind waiting 3 minutes or more to get a fun fight though. I would much rather the guys who would do more good for the team by sitting in our MCC and not dying 20 times just stop getting my team cloned. The Amarr scout bonus is like the old Amarr sentinel bonus. No one needed 25% reduction to overheat damage on a heavy;_;
Seymour KrelbornX Holdfast Syndicate Amarr Empire 416	Posted - 2014.07.10 01:32:00 - [18] - Quote Kevall for CPM1! a very amusing and vital thread. a better matchmaking system would at most least double the fun factor of dust for many.... there must be a way, even with a low pop game.
Kevall Longstride DUST University Ivy League 1470	Posted - 2014.07.10 08:29:00 - [19] - Quote Seymour KrelbornX wrote: Kevall for CPM1! a very amusing and vital thread. a better matchmaking system would at most least double the fun factor of dust for many.... there must be a way, even with a low pop game. Thanks for that. I actually put matchmaking above the NPE as my feature to add for Dust 514 for that very reason. The NPE would need a lot of work and a client update to be fit for purpose. But a matchmaking fix is all server side, comparatively quicker to do and would help all players not just the new guys. CPM1 Candidate CEO of DUST University


Pages: 1 :: [one page]