Public Test - 24/jun/2014
#161
Posted 24 June 2014 - 08:07 PM
#162
Posted 24 June 2014 - 08:23 PM
#163
Posted 24 June 2014 - 09:26 PM
Sandpit, on 24 June 2014 - 08:23 PM, said:
Russ said here http://mwomercs.com/...sizes-and-more/
Quote
So I guess sadly for the moment that is a no.
But it would be a good idea to sort out group matchmaking, would have to make it worthwhile to solo people tho, like a cbill/exp bonus or something for dropping into the premade queues.
#166
Posted 24 June 2014 - 10:09 PM
"Not authorized"
#167
Posted 24 June 2014 - 10:13 PM
Sky Hawk, on 24 June 2014 - 07:19 PM, said:
Not quite. The old matchmaker had request starvation issues. It was incapable of examining all potential matches to see if a game could be produced. It's difficult to explain without going into a lot of algorithmic details, but we were looking for a lot more than a success or failure result. We needed to see how the new system behaved with users adapting to the weight class queue constraints and the new group mechanics, and how that affected the new matching algorithms.
#168
Posted 24 June 2014 - 10:22 PM
Deathlike, on 24 June 2014 - 07:21 PM, said:
There was no turning off of the new matchmaker. The matchmaker currently on production is entirely deleted from public test in fact.
The only change made was to disable the strict weight class matching requirement, since our group queue was so low in population that picking the first 24 players to queue as groups was still resulting in excessively high wait times. The 3's rule was still in place with release valves, one pre-made per team for non-group games was still in place, 5+ sized groups were still in place. I believe strict weight class matching was disabled at around 6 PM pacific, but I did not record the exact time. Our recorded telemetry contains exact audits of team composition, so we can easily identify which games were produced under which rule sets.
Disabling the strict weight class for weight class rule was a compromise driven purely by population sizes.
#169
Posted 24 June 2014 - 10:29 PM
Karl Berg, on 24 June 2014 - 10:13 PM, said:
But you already know players wont adapt, thats why you programmed the new MM to allow more than 3 chassis of the same class into game when a certain time ticks by right?.
If you believed players would adapt you would not of added that feature into the MM.
So Karl tell me, if your ELO system works, why not have the MM match teams via an ELO + tonnage system, i mean why doesnt (roughly same skill level) + (roughly same hardware tonnage) work? letting people play the chassis they want?
Karl Berg, on 24 June 2014 - 10:22 PM, said:
The 3's rule was still in place with release valves,
So in fact its there but isnt, within a certain time limit, the MM with ignore the 3s rule and carry on to allow a game with 8 Assaults on each team, effectively giving an MM that works on a tonnage/class limit
Edited by N0MAD, 24 June 2014 - 10:48 PM.
#170
Posted 24 June 2014 - 10:34 PM
MonkeyCheese, on 24 June 2014 - 07:36 PM, said:
http://mwomercs.com/...ost__p__3504038
Now I know the test population was really low, but if you are trying to do a rule of 3/ 3x4 matchmaker then why the heck can one weight class get to 8?
This is due to the release valves kicking in. The matchmaker will try it's best to produce a 3's game, but it will ease off as people in the queue wait for too long. All of these parameters are adjustable by us; how long someone should wait before relaxing three's, how long should they wait before relaxing Elo, how rapidly should Elo or threes be relaxed, etc.
This allows the system to adapt to a wide variety of situations, such as the tiny queue sizes we saw here on PTS, or sales and events involving popular mechs which we've seen can distort queue composition dramatically.
#171
Posted 24 June 2014 - 10:40 PM
N0MAD, on 24 June 2014 - 10:29 PM, said:
If you believed players would adapt you would not of added that feature into the MM.
Actually I simply can't predict how players will adapt, or if they will adapt to an exact 25% per weight class queue composition. Any imbalance in queue composition results in wait queues, and so if we want to restrict matching to such specific constraints that are ultimately under player control, we must implement the system to handle all input cases. Without the ability to handle all input cases, whether required or not, the algorithm would simply not be robust.
With our specific solution, any queue imbalance results in a set of games being launched under a different ruleset. Once the imbalanced queues are emptied, and wait times are back within tolerance levels, the matchmaker resumes producing fully constrained games.
#172
Posted 24 June 2014 - 10:41 PM
Karl Berg, on 24 June 2014 - 10:22 PM, said:
There was no turning off of the new matchmaker. The matchmaker currently on production is entirely deleted from public test in fact.
The only change made was to disable the strict weight class matching requirement, since our group queue was so low in population that picking the first 24 players to queue as groups was still resulting in excessively high wait times. The 3's rule was still in place with release valves, one pre-made per team for non-group games was still in place, 5+ sized groups were still in place. I believe strict weight class matching was disabled at around 6 PM pacific, but I did not record the exact time. Our recorded telemetry contains exact audits of team composition, so we can easily identify which games were produced under which rule sets.
Disabling the strict weight class for weight class rule was a compromise driven purely by population sizes.
My suggestion was posted in the feedback section of the PTS (MM):
http://mwomercs.com/...ost__p__3504845
My thinking is that if you changed the 5-10 man team building constraints/requirements (for instance, forcing 2/2/2/2 team construction for 5 to 8-mans), you might be able to produce a better result. What you will need to also do is add in "opt-in" solo queuing to achieve some of the means. That would help relieve the MM's job a bit as well.
Edited by Deathlike, 24 June 2014 - 10:43 PM.
#173
Posted 24 June 2014 - 10:45 PM
Ripper X, on 24 June 2014 - 08:07 PM, said:
Yes, this is something that will need to be considered carefully in future, and I will definitely pass on this feedback; thanks!
Windsaw, on 24 June 2014 - 10:09 PM, said:
"Not authorized"
Sorry, PTS was disabled at 8 PM pacific.
#174
Posted 24 June 2014 - 11:03 PM
Sandpit, on 24 June 2014 - 08:23 PM, said:
I would need to clarify with design and product owners, but it seems like something that could be considered in future if there appeared to be enough interest.
Obviously one of our goals was to simply the number of factors you have to deal with to queue up and get in game. There are many more changes still to come intended to help streamline things even further.
#175
Posted 24 June 2014 - 11:05 PM
Edited by Ductus Hase, 24 June 2014 - 11:06 PM.
#176
Posted 24 June 2014 - 11:09 PM
Karl Berg, on 24 June 2014 - 10:34 PM, said:
Like when you release a spider hero mech and everyone starts driving them and throws the machmaker balance out the window right? hint hint spider hero!
Also when urban mechs come you better prepare your solo queue matchmaker for 12v12 urbanmech battles!!!
#177
Posted 24 June 2014 - 11:10 PM
Karl Berg, on 24 June 2014 - 10:40 PM, said:
With our specific solution, any queue imbalance results in a set of games being launched under a different ruleset. Once the imbalanced queues are emptied, and wait times are back within tolerance levels, the matchmaker resumes producing fully constrained games.
So the logical question here.
So if the imbalanced queues doesnt empty, which is quite likely as numbers constantly show around 70% of players prefer Heavy/Assaults, what then? constant games of 8+ heavy/assault mechs per team per game?.
#178
Posted 24 June 2014 - 11:24 PM
N0MAD, on 24 June 2014 - 11:10 PM, said:
So if the imbalanced queues doesnt empty, which is quite likely as numbers constantly show around 70% of players prefer Heavy/Assaults, what then? constant games of 8+ heavy/assault mechs per team per game?.
I really dont think it would be that bad on the live server, they said the test server was like 10% of the turn out for the clans test server.
The live server has so so so many more players than that small fragment we seen on the test server. I think the system will easily be able to find 24 people before the matchmaker has to let up on the weight class restrictions and things like 8 assaults or 10 lights on a team happen. I assume they will be rare.
Once all the bugs are squashed and the new system is live that is.
#179
Posted 24 June 2014 - 11:32 PM
MonkeyCheese, on 24 June 2014 - 11:09 PM, said:
Like when you release a spider hero mech and everyone starts driving them and throws the machmaker balance out the window right? hint hint spider hero!
Also when urban mechs come you better prepare your solo queue matchmaker for 12v12 urbanmech battles!!!
Yes exactly. I suggested they keep the urban in reserve as a PTS only mech to draw users. I was laughed at
#180
Posted 24 June 2014 - 11:36 PM
N0MAD, on 24 June 2014 - 11:10 PM, said:
So if the imbalanced queues doesnt empty, which is quite likely as numbers constantly show around 70% of players prefer Heavy/Assaults, what then? constant games of 8+ heavy/assault mechs per team per game?.
Even under these conditions, nope. We'd be able to produce some percentage of threes games, with several assault / heavy dominated games to empty the queue of the remaining users. The exact percentages would be a function of the tolerance and decay values we use in the matching algorithm.
7 user(s) are reading this topic
0 members, 7 guests, 0 anonymous users