Jump to content

Paging Karl Berg...karl Berg, Please Pick Up The White Courtesy Phone...


1911 replies to this topic

#1301 One of Little Harmony

    Member

  • PipPipPipPipPip
  • The Devoted
  • The Devoted
  • 159 posts

Posted 19 July 2014 - 11:23 AM

View PostVoid Angel, on 19 July 2014 - 08:29 AM, said:

I'm unclear how you can know what probability you have of being in the last slot when you don't know your Elo, others' Elo, or (as JZ suggested) how often players of various Elo ranks actually play?


Well,it was in a parenthetical because it wasn't really necessary to what I was saying, but I guess it was confusing because I had an ellipsis in there where I just gathered people would be able to infer my meaning from the previous discussion and I didn't want to type a whole paragraph explaining if I didn't have to.

Assuming the null hypothesis, that the order of ranking in lances is not influenced by Elo, and the second condition of only solo drops, one has a 1/12th chance of being in any given slot. However, in current play, one can drop with up to 1 group on their team, so conservatively you will estimate your chance of having a given slot as 1/8, because, while the value is probably less than 1/8 chance as you sometimes have a 2man, 3man or no premade on your team, if you can prove this highest condition the other ones are trivial. Thus, the probability with the null hypothesis of dropping in the last slot is 1/8th, or in the last two slots is 1/4th when one assumes a 4man premade on one's team.

#1302 p4r4g0n

    Member

  • PipPipPipPipPipPipPipPip
  • Knight Errant
  • 1,511 posts
  • LocationMalaysia

Posted 19 July 2014 - 11:43 AM

@One of Little Harmony

AFAIK, the solo queue is now purely a solo queue with no groups at all. There have been a number of recent threads referencing a tweet to this effect by Russ Bullock.

#1303 Not Bob

    Member

  • PipPipPipPipPipPipPip
  • CS 2022 Referee
  • CS 2022 Referee
  • 688 posts
  • LocationThat one place

Posted 19 July 2014 - 11:53 AM

View Postp4r4g0n, on 19 July 2014 - 11:43 AM, said:

@One of Little Harmony

AFAIK, the solo queue is now purely a solo queue with no groups at all. There have been a number of recent threads referencing a tweet to this effect by Russ Bullock.



Sorry, but Solo que is pugs with one group of 2-4 max (On each time) :)

It was stated in the patch noes, I do believe


Edit: Group que is all groups 2-10 or 12 - With 12s having priority being matched against each other first, then other groups

Edited by Not Bob, 19 July 2014 - 11:54 AM.


#1304 Cimarb

    Member

  • PipPipPipPipPipPipPipPipPip
  • Caladbolg
  • Caladbolg
  • 3,912 posts
  • Twitter: Link
  • Twitch: Link
  • LocationA hop, skip and jump from Terra

Posted 19 July 2014 - 12:06 PM

View PostNot Bob, on 19 July 2014 - 11:53 AM, said:



Sorry, but Solo que is pugs with one group of 2-4 max (On each time) :)

It was stated in the patch noes, I do believe

Edit: Group que is all groups 2-10 or 12 - With 12s having priority being matched against each other first, then other groups

That was the goal originally, yes, but Russ stated on Twitter that it is actually SOLO-ONLY, and would stay that way indefinitely, as it was giving solo players a "very good experience".

#1305 Not Bob

    Member

  • PipPipPipPipPipPipPip
  • CS 2022 Referee
  • CS 2022 Referee
  • 688 posts
  • LocationThat one place

Posted 19 July 2014 - 12:53 PM

View PostCimarb, on 19 July 2014 - 12:06 PM, said:

That was the goal originally, yes, but Russ stated on Twitter that it is actually SOLO-ONLY, and would stay that way indefinitely, as it was giving solo players a "very good experience".



Oh.. Well, I could have sworn when I was running some 2 mans and 3 mans last night we were in the solo que, but ah well. Guess I didn't get that recent info lol :)

#1306 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 01:27 PM

View PostKyle Reece, on 16 July 2014 - 09:51 AM, said:

Karl, not wanting to put words in your mouth but surely in regards to how the MWO implementation of Elo works wouldn't that really be...

Or have I interpreted the various CC and dev posts incorrectly?


Correct Kyle; for our specific implementation in MWO, team outcome is the determining factor for Elo adjustment.

#1307 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,961 posts
  • LocationParanoiaville

Posted 19 July 2014 - 01:34 PM

Currently the solo queue is only solo players - don't feel bad; the devs, in trying to get more information out to us, often put little things like that in twitter feeds or whatnot and forget to announce it in other places.

View PostOne of Little Harmony, on 19 July 2014 - 11:23 AM, said:

Assuming the null hypothesis, that the order of ranking in lances is not influenced by Elo, and the second condition of only solo drops, one has a 1/12th chance of being in any given slot. However, in current play, one can drop with up to 1 group on their team, so conservatively you will estimate your chance of having a given slot as 1/8, because, while the value is probably less than 1/8 chance as you sometimes have a 2man, 3man or no premade on your team, if you can prove this highest condition the other ones are trivial. Thus, the probability with the null hypothesis of dropping in the last slot is 1/8th, or in the last two slots is 1/4th when one assumes a 4man premade on one's team.

Good analysis!

My only concern is that we don't know enough about the matchmaker to determine whether differences in our lance assignment are due to Elo, player composition during the hours we play, or some other factor. If Elo does somehow affect lance assignment, not knowing how it does so can make things difficult. For example, the matchmaker could take the three highest-Elo random players and assign each of them to a different lance, then repeat; it could assign players in order of Elo - or it could simulate a game of Duck-Duck-Goose. It probably doesn't do it randomly, but it could play The Name Game with our Mechwarrior handles and assign a numerical value to the result, for all we know.

Edited by Void Angel, 19 July 2014 - 01:34 PM.


#1308 SnagaDance

    Member

  • PipPipPipPipPipPipPipPip
  • Little Helper
  • Little Helper
  • 1,860 posts
  • LocationThe Netherlands

Posted 19 July 2014 - 01:45 PM

Had an interesting thing just now. Me and my buddy were a 2-man. Dropped and got deployed in 2 different lances. No company leader re-shuffle or anything. Seems like smaller groups are broken up (within the same team) to help fill up larger groups?

#1309 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 01:49 PM

View PostCimarb, on 16 July 2014 - 11:56 AM, said:

Your edits are exactly why I dislike this system... Having my personal "skill" adjusted based upon team performance irritates me, but I have already argued that way too much for this thread...


Try not to worry about it. It is precisely for this reason that we keep Elo as a hidden variable. Trying to bias specific users Elo adjustments despite your teams performance results in weird edge cases that we can't possibly resolve.

If the system has incorrectly calculated your skill, you will on average be placed into games where you will affect the outcome of the match much more than expected. This eventually leads to the system correcting itself.

#1310 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 19 July 2014 - 02:01 PM

View PostSnagaDance, on 19 July 2014 - 01:45 PM, said:

Had an interesting thing just now. Me and my buddy were a 2-man. Dropped and got deployed in 2 different lances. No company leader re-shuffle or anything. Seems like smaller groups are broken up (within the same team) to help fill up larger groups?

Yes. This behavior is fairly obvious: How else do you deal with 3 man groups, for example?

#1311 Wintersdark

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • 13,375 posts
  • Google+: Link
  • Twitter: Link
  • LocationCalgary, AB

Posted 19 July 2014 - 02:16 PM

View PostVoid Angel, on 19 July 2014 - 08:29 AM, said:

I'm unclear how you can know what probability you have of being in the last slot when you don't know your Elo, others' Elo, or (as JZ suggested) how often players of various Elo ranks actually play?


I've always explained it by pointing out that low-range Elo players are often there because of bad teamwork practices - like being hate-filled rage monkeys - in addition to bad player skills. Ironically, that means that the stereotypical abusive player often really is better than his ranking (and therefore his average teammates) would suggest - it's just that he's so abusive and hostile whenever anything goes wrong that he sabotages any team he's on.

Once you get far enough down in Elo (say by, er, raging back at the previously mentioned jerks) you start getting matched with players who are so uncooperative and rude that your own performance feels futile. As an example, once you get to 1200-ish Elo in League of Legends, you used to get a ton of people who would threaten to "feed" (that is, deliberately allow the enemy to kill them) if the rest of the team wouldn't allow them the spot on the team they wanted, or would be AFK from the start of the match, or would deliberately switch to a different spot whose player had already locked in, then claim that they were the one who locked in first during the match (chat logs are only available in-match.) I had one troll who I would love to do redactable things to, even to this day; he played the "I'm switching" game, and then followed people around to screw with them - all the while being scrupulously polite and friendly in chat.

This is Elo Hell. Sure, you'll eventually get out of it if you play enough - but can you play enough, and still live with what's left of your soul?

Interestingly, this happens in MWO as well.

I both deliberately and unintentionally tanked my Light elo some months ago, when the Spider was released. I hadn't had a Light since looooong before stats where recorded at all, so my Light Elo didn't exist. I loaded up a Spider with machine guns and flamers (note, this was before the first of the MG fixes, when they were truly horrible) and worked up a smashing 0.3KDR with them and 0.4 WLR. I kept at it for a long time, because I was interested in how the game worked at lower Elo rankings (given how different it is at mid to high Elo bands).

And what you say above? It's absolutely the case. So many people who are absolutely hostile. it's not that the player skills get worse than averagish because almost everyone can aim and shoot at things in a video game to a mediocre level of ability at the minimum. But you get people there who are really terrible to play with, and worse...

It's extremely hard to get out once you are there. I hadn't anticipated this, as I don't play any other PVP games (I hate PvP overall, I just love Mechwarrior) and had no experience with such systems. I'm a bad light pilot, but I'm an experienced pilot overall pushing 5000 drops, I know how to put a mech together and play at at least an average level... But once you're down there, games are almost completely random. It's so easy to lose matches because the people on your team are not bad (or are not JUST bad) but are actually sabotaging your team deliberately.

Teamwork is everything.

#1312 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 02:16 PM

View PostSnagaDance, on 17 July 2014 - 12:24 AM, said:

Hi Karl. Got a question you'll probably need to ask others for the answer but I hope you can help: "How does the game determine in what Lance you start (in PUGs)?" These last 6 months I've consistently dropped in Alpha lance, hardly any exceptions except when a pre-made (who I know always gets in Alpha) was included. And since yesterday I'm suddenly a lot in Charlie, plus a couple of times in Beta and just a little Alpha? I'm not complaining, it's a nice change for now, but I'd like to know the mechanism behind this. (sorry for the little block of text, I'm behind a work computer that doesn't allow me to use Enter to get a new line on these forums for some reason).


Yup, this has been slightly modified for a few reasons. The old algorithm is pretty much as I described earlier in this thread.

There was never any consideration of Elo in lance assignment, it has always been driven off of group size. This is due to the requirement that we split groups as infrequently as possible.

Previously, this assignment of players to lances was done on the dedicated server. Last patch, I deleted this dedicated server code, and re-implemented assignment to happen on the matchmaker itself. This resolves a potential bug, where the dedicated server could clobber lance assignment done in private lobbies.

The new algorithm on the matchmaker is simply this:
- Sort requests by group size
- Select a random seed lance
- For each request
   - Increment seed lance
   - For each user in request
	  - While seed lance is full
		 - Increment seed lance
	  - End
	  - Place user into seed lance
   - End
- End


This correctly handles all manners of input group sizes with very little code. Randomizing the seed lance means the largest group is no longer assigned to Alpha lance 100% of the time. Hopefully now you will see a much larger variety of initial spawn points in game if you tend to play with a specific group size. Also, no more bias against Charlie lance. :)

#1313 Goose

    Member

  • PipPipPipPipPipPipPipPipPip
  • Civil Servant
  • Civil Servant
  • 3,463 posts
  • Twitch: Link
  • LocationThat flattop, up the well, overhead

Posted 19 July 2014 - 02:20 PM

View PostKarl Berg, on 19 July 2014 - 01:49 PM, said:

Try not to worry about it. It is precisely for this reason that we keep Elo as a hidden variable. Trying to bias specific users Elo adjustments despite your teams performance results in weird edge cases that we can't possibly resolve.

If the system has incorrectly calculated your skill, you will on average be placed into games where you will affect the outcome of the match much more than expected. This eventually leads to the system correcting itself.

The system is a trailing indicator, and inherently noisy: This is good, in the "reach should exceed grasp" meaning, as otherwise matches could get stratified 'n boring. Posted Image

#1314 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 02:27 PM

View PostJman5, on 17 July 2014 - 06:38 PM, said:

Hey Karl, in the past few weeks, I've noticed some occasional, but severe hit registration problems in the private lobby games. Most games are fine, but every once in a while I'll get a match where everyone is experiencing bad hit reg.

Normal matches are fine though, so it's very strange. For the record I'd like to think I'm not the kind of person who blames hit reg every time he misses.

I'm just wondering if you guys are aware of anything going on with private lobby servers or connections.


That's interesting.

It might be related to the current 'lag' issues some games are experiencing on production. There are known repro steps for the lag issue that some of the competitive players brought to our attention some time ago. They were able to reproduce these failure conditions for us in a controlled environment. Unfortunately these behaviours are causing the underlying CryNetwork system on the dedicated server to enter a negative performance feedback cycle. The server is essentially unable to keep up with the load being placed on it in these circumstances.

We've tested one set of fixes with the players who brought this to our attention already. This test unfortunately didn't succeed. We have a more aggressive set of changes in the works that we will get into test as soon as we possibly can.

#1315 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 02:30 PM

View PostTheCaptainJZ, on 17 July 2014 - 09:36 PM, said:

No, the ready screen when you hit the ready button in match. already loaded the match. if it's not the ready screen i don't know what you call it


Oh, are you referring to ordering within a lance once a match starts? That specific display ordering is handled by the HUD.

View PostTheCaptainJZ, on 18 July 2014 - 11:44 PM, said:

You could have something like this:
ALPHA Player A
Player B
Player C
Player D

but to a teammate on voice comms that you are grouped with, it could show as
ALPHA Player B
Player D
Player A
Player C


edit: I see, that was what you were referring to. I think ordering within a lance is essentially random. It's most likely affected by entity spawn ordering on your client at game start.

#1316 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 19 July 2014 - 02:38 PM

View Postp4r4g0n, on 19 July 2014 - 11:43 AM, said:

@One of Little Harmony

AFAIK, the solo queue is now purely a solo queue with no groups at all. There have been a number of recent threads referencing a tweet to this effect by Russ Bullock.


Yes, I can confirm this. Solo queue is 100% solo only, and has been almost since we released the new matchmaker.

Groups all play in group queue; and since the last patch we've added some pretty heavy biases to force similarly sized groups together.

#1317 Jman5

    Member

  • PipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 4,914 posts

Posted 19 July 2014 - 03:15 PM

So did you guys enjoy watching the Clan vs IS test matches? Any fun facts you care to share with us?

#1318 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,961 posts
  • LocationParanoiaville

Posted 19 July 2014 - 03:26 PM

View PostKarl Berg, on 19 July 2014 - 02:16 PM, said:


Yup, this has been slightly modified for a few reasons. The old algorithm is pretty much as I described earlier in this thread.

There was never any consideration of Elo in lance assignment, it has always been driven off of group size. This is due to the requirement that we split groups as infrequently as possible.

Previously, this assignment of players to lances was done on the dedicated server. Last patch, I deleted this dedicated server code, and re-implemented assignment to happen on the matchmaker itself. This resolves a potential bug, where the dedicated server could clobber lance assignment done in private lobbies.

The new algorithm on the matchmaker is simply this:
- Sort requests by group size
- Select a random seed lance
- For each request
   - Increment seed lance
   - For each user in request
	  - While seed lance is full
		 - Increment seed lance
	  - End
	  - Place user into seed lance
   - End
- End


This correctly handles all manners of input group sizes with very little code. Randomizing the seed lance means the largest group is no longer assigned to Alpha lance 100% of the time. Hopefully now you will see a much larger variety of initial spawn points in game if you tend to play with a specific group size. Also, no more bias against Charlie lance. :)

Good to know;

But what about lance order within an established group? I haven't been doing 12-mans for a while, but when I was we noticed that it was impossible for us to organize our lance teams with our own team makeup in mind. For example, if we wanted all of the guys who were performing the scout role to be organized into Charlie Lance, there was no way to do that. We fiddled with the system to see if we could find a (non-exploit) way to get that done, but to no avail.

Edited by Void Angel, 19 July 2014 - 03:27 PM.


#1319 Cimarb

    Member

  • PipPipPipPipPipPipPipPipPip
  • Caladbolg
  • Caladbolg
  • 3,912 posts
  • Twitter: Link
  • Twitch: Link
  • LocationA hop, skip and jump from Terra

Posted 19 July 2014 - 05:24 PM

View PostKarl Berg, on 19 July 2014 - 01:49 PM, said:


Try not to worry about it. It is precisely for this reason that we keep Elo as a hidden variable. Trying to bias specific users Elo adjustments despite your teams performance results in weird edge cases that we can't possibly resolve.

If the system has incorrectly calculated your skill, you will on average be placed into games where you will affect the outcome of the match much more than expected. This eventually leads to the system correcting itself.

I appreciate that, and I am not worried about my specific Elo. I am just worried about user experience, and when it seems like Elo does not even work because of the weird and seemingly random collaboration of people in a particular match, it is a very poor experience for both ends of the Elo spectrum, as well as those in between.

Just please continue to work on it and I am sure it will work out, but I would love to have "good" matches individually rather than horrible matches "averaged" over six months to look good on a stat sheet.

#1320 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,961 posts
  • LocationParanoiaville

Posted 19 July 2014 - 06:42 PM

That's not what Elo does, as you have repeatedly been told.





43 user(s) are reading this topic

0 members, 43 guests, 0 anonymous users