Jump to content

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


1911 replies to this topic

#201 Peter2000

    Member

  • PipPipPipPipPipPip
  • 269 posts

Posted 12 April 2014 - 12:55 PM

Regarding some of the Teams/Matchmaking debate:

Two problems come to mind. One easy, one hard.

The easy problem is that while pairing together might generally be stronger than solo play, pairing together might not confer the same ELO benefit in all ELO ranges. For example, premades are a recognizable power at top ELO (you can see who they are, since the population is smaller, and good premades can carry REALLY hard). However, at lower ELO, game knowledge and coordination are lower, and even if you coordinate, if you can't hit the broadside of a barn, it matters little. Disclaimer: these numbers (and all that follow on ELO) are pulled out of my Rear Center Torso. The effect might actually be the opposite direction in practice - I'm just providing an example of the fact that the impact of grouping might not be universally the same.

So, the answer might be to (based on empirical information you have access to that we don't), derive the ELO boost that a group gives, not as a universal constant, but as a function of player ELO. E.g. Grouped_4_ELO = 50 + (Base_ELO*1.1). It doesn't even need to follow a determined function - you could use smoothing/interpolation to follow an arbitrary empirically derived pattern.

The hard problem is to account for systematic differences in group play. Some groups may be very coordinated and able to significantly outperform their theoretical ELO rating, because their teamplay is quite good. Others might not even be on comms. The average-by-ELO method described above might not capture this. At the same time, until there is a lot of data on a player in this regard, it is doesn't make sense to treat him as far from average.

Therefore it might make sense to treat players as an average (as above) until they have ~100 premade games and as many pure PUG under their belt, then compare their performance in PUG relatative to group, and use that to form a differential modifier for group play for each individual. E.G. Player X has played 100 PUG and 100 4 mans. His ELO currently is 1000. Up until this point, his ELO modifier for group matches was +150 - the default value for 1000 ELO. He played all "fair" matches with ~50% win expectancy, and won 50% of his PUG matches, and 60% of groups. His ELO now gets a +100 adjustment, whenever he is grouped.



Regarding the SHS/DHS tonnage/crit tradeoff:

The traditional argument for SHS being relevant is that they consume fewer crit slots than DHS, therefore making them balanced. While this seem plausible on the surface, it is a fallacy.

Why? EHS - Effective Heat Sinks - the total level of cooling is totally skewed in DHS favor. If you run SHS, your EHS = your SHS. If you run DHS, your EHS = 2.0xEngine DHS + 1.4xOtherDHS.

If you take DHS and at least a 250-rated engine (basically all 'Mechs use an engine this large, or larger), and NO heat sinks, you have already saved 10 crit slots compared to SHS cooling the same quantity (same EHS for 10 in-engine doubles and 20 singles). Excluding engine slots, it takes carrying nearly 30 (I have a table for the exact numbers of crits saved) SHS before you're saving ANY crit slots compared to an eqivalently cooled DHS 'Mech (and you're paying upwards of 12 tons for those one or two slots!). The bigger the engine gets, the higher the tradeoff point.

Since you get so much for free out of engine DHS, SHS need to do something better than them to be balanced. Like heat threshold.

Edited by Peter2000, 12 April 2014 - 02:05 PM.


#202 Peter2000

    Member

  • PipPipPipPipPipPip
  • 269 posts

Posted 12 April 2014 - 01:01 PM

Also, P.S. this thread is awesome, thanks for doing this.

#203 aniviron

    Member

  • PipPipPipPipPipPipPipPipPip
  • 3,752 posts

Posted 12 April 2014 - 01:14 PM

Karl, a huge thanks for all your words- they really do make a difference, and give me some hope for the future of the game again.

But it's Saturday! If you're not in the office, take a well-deserved rest.

Edited by aniviron, 12 April 2014 - 01:16 PM.


#204 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 12 April 2014 - 01:34 PM

I have to agree on a group Elo modifier not being the greatest idea. As stated before, I see both good groups carrying hard, and bad groups absolutely tanking matches. The effect is multiplied by the 4-man moving together, but it is not always a positive effect. Ideally, group and solo drops should have separate Elo scores to reflect that.

The same goes for tracking Elo per mech or even variant rather than weight class. There are extreme differences between mechs within a weight class, and you can never predict what a player will switch to next. The score becomes a range instead of a strictly converged value. Multiply for 24 players, and you get a major random factor.

#205 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 12 April 2014 - 02:42 PM

View PostHammerhai, on 12 April 2014 - 08:31 AM, said:

So here my question: Specifically, can we look forward to a Command wheel type interface or a bug fix on that? Where IS VOIP/ Command wheel on the bug fix/priority list?


At this point, this is all side-work. We're not to delay any of the big items with this work (Clans, CW, some others not yet discussed). Russ cleared us for VOIP work, Brian Buckton is looking at our implementation options.

#206 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 12 April 2014 - 02:53 PM

View PostSandpit, on 12 April 2014 - 09:51 AM, said:

In regards to ideas for groups:


Some excellent ideas there. I think you'll be pleased within the next few weeks.

View PostPeter2000, on 12 April 2014 - 12:55 PM, said:

Regarding some of the Teams/Matchmaking debate:


View PostModo44, on 12 April 2014 - 01:34 PM, said:

I have to agree on a group Elo modifier not being the greatest idea.


Here as well.

Any professional statisticians out there? :)

#207 Hatchet Jack

    Member

  • PipPipPip
  • Overlord
  • Overlord
  • 56 posts
  • LocationCalgary, Alberta

Posted 12 April 2014 - 03:12 PM

View PostKarl Berg, on 12 April 2014 - 02:53 PM, said:


Some excellent ideas there. I think you'll be pleased within the next few weeks.




Well,
Karl, you're knocking the ball out of the park here the past few days. Not to mention a weekend update from Matt.
As for getting the community involved...

Quote

Here as well.

Any professional statisticians out there? :)

Posted Image

#208 Roadbeer

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Elite Founder
  • 8,160 posts
  • LocationWazan, Zion Cluster

Posted 12 April 2014 - 03:14 PM

View PostKarl Berg, on 12 April 2014 - 02:53 PM, said:

Any professional statisticians out there? :)

Isn't that the job one takes whey they fail Algebra I at a Junior College?

#209 Heffay

    Rum Runner

  • PipPipPipPipPipPipPipPipPipPip
  • The Referee
  • The Referee
  • 6,458 posts
  • LocationPHX

Posted 12 April 2014 - 03:17 PM

View PostRoadbeer, on 12 April 2014 - 03:14 PM, said:

Isn't that the job one takes whey they fail Algebra I at a Junior College?


No, I think it's someone who posts a poll on the forum to try to figure out what the people who play the game want. :)

#210 Roadbeer

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • Elite Founder
  • 8,160 posts
  • LocationWazan, Zion Cluster

Posted 12 April 2014 - 03:20 PM

View PostHeffay, on 12 April 2014 - 03:17 PM, said:


No, I think it's someone who posts a poll on the forum to try to figure out what the people who play the game want. :)

Oh, then the forum has about 1,000 of them, just go to Feature Suggestion sub-forum. Or we could get Victor back...

Do you want 3PV
No
Hell No

Unbiased poll is unbiased.

#211 B0oN

    Member

  • PipPipPipPipPipPipPipPipPip
  • 2,870 posts

Posted 12 April 2014 - 04:11 PM

About VoIP :
Is it feasible to integrate a VoIP system into the game that takes into account wave-travel-impediments (hills,trees,steel structures,etc...) and positional placement of sending/receiving units ?

One possible example would be the ACRE TS plugin which connected ArmA2/ArmA3 to the used TS server and provided spatial audio positioning of users (i.e. be able to hear someone talking according to distance from speaker| ~audible radius with proper fall-off of 20m) and futhrmore implemented real-world impediments for wavetravelling properties of also implemented radios (i.e.: hand-radius for squadlevel chat with a range of 1km; backpack radios with up to 2,5k range; vehicle mounted, massively amped radiostations with range of up to 10k) .

So, if one was talking closeby you could hear him directly and correctly positioned (up, dwn, left, right), if he was 200m away use that squadradio (if you knew the chnnel hewas listening to), if he was 3k away you had to have either a backpack radio plus a vantage point free of obstructions/flat plain or a powerful, vehicle-mounted fixed radio.

Thats it for now

#212 juju2112

    Member

  • Pip
  • Overlord
  • Overlord
  • 16 posts

Posted 12 April 2014 - 04:14 PM

Thanks for the reply, Karl. That's interesting!

Regarding this point here:

View PostKarl Berg, on 12 April 2014 - 10:56 AM, said:

  • When a bug is fixed, is a unit test created to test for it in future builds?
Sometimes, when it is possible to. Most usually not however, as most bugs require specific edge cases to be set up in a database beforehand, or require some specific interaction of multiple servers to manifest. This is related to the functional testing above.



Where I work, some areas are using CA Lisa Service Virtualization for this. So, you can mock up a fake database, web service, or whatever, and your unit tests hit the Lisa instance for those things instead. It's a good way to test things like that that are inherently integrated into other systems. I certainly haven't had the time to learn it myself, though. Too busy, much like yourself I presume!

Edited by juju2112, 12 April 2014 - 04:14 PM.


#213 o0cipher0o

    Member

  • PipPipPipPipPipPip
  • FP Veteran - Beta 1
  • FP Veteran - Beta 1
  • 353 posts
  • LocationItaly

Posted 12 April 2014 - 04:39 PM

View PostKlappspaten, on 12 April 2014 - 09:13 AM, said:


Let me guide your attention to post #145 in this thread, its about the question you jjust asked.

http://mwomercs.com/...ost__p__3287064


Well, i did read that post, and also the answer to it. What i'm talking about, aren't the rendering issues related to low settings and distances, but the terrain hitboxes. For example, in tourmaline, in the center, the two cristals walls right on the top of e5, have the hitboxes protruding slightly over the actuall terrain geometry, so that, even from just a bunch of meters away, when everything is rendered, if you shoot at a mech you can see just the head, for example, you'll end hitting an invisible wall, whereas the projectiles should have been fully capable of reaching The target.
I'm not talking about the appearing/disappearing objects around the maps, but about the hitboxes that in some points doesn't coincide with the actual geometry of the map.

#214 MischiefSC

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPip
  • The Benefactor
  • The Benefactor
  • 16,697 posts

Posted 12 April 2014 - 04:43 PM

View PostKarl Berg, on 12 April 2014 - 02:53 PM, said:

Any professional statisticians out there? :)


Not on weekends and only if you provide the software that does all the hard work for me :D.Technology has made me lazy; I don't even have to remember phone numbers anymore, not to mention how to calculate margin of error. The big secret? You don't need to know anything about statistics or analytics to be good at it. You just have to know what stuff is called so that you can pick the right function in the spreadsheet.

It actually wouldn't be a hard variable to solve for. Playing in a premade doesn't necessarily given an advantage; what you're really wanting to measure is synergy. How well specific players do together.

So you'd track a paired Elo for every player who plays with another player, this way it self-corrects for the size of the premade team. They get an Elo modifier based on the difference between their 'stock' Elo and their premade Elo with that player and average it with everyone in that specific premade instance. You'd also want a minimum seed threshold before you applied the variable. Dunno what the standard deviation is for Elo in MW:O

So player A has an Elo of 1600. He plays with player B who has an Elo of 1800 and player C who has an Elo of 1200 (he loves his steering wheel, what can he say). They have estimated Elo of 1533 together. HOWEVER, they use VoIP and C plays a lot better when he's with B but his builds don't work well with As builds and A doesn't really go out of the way to help him, while A plays about the same regardless of playing with both. B really feeds off of A and C and does even better with either.

The result would be that as a team of 3 A gets no real modifier, B does 8% better with just C and 5% better with just A. C gets 0 modifer for A and 20% for B.

Result would be as follows:

A+B+C in a team - A=1600, B=2016, C= 1440, so their new estimated Elo would be 1685.

A+B = 1600+1890/2= 1745

Etc.

Make sense? It would involve tracking Elo not just for each player but for each player relative to the people they drop in premade with. While down the road, premade or not tracking relative synergy for specific players (especially if you ever decide to drill down on things like specific chassis and variants, loadouts, etc) would be a great way to get a very precise Elo for each player....

The reality though is that even in a statistically big change like A,B,C had (over 200 points swing when playing together) it would still keep them in the 1500+ bucket and so no significant change.

If players in a premade are good enough together to carry at 1500 or greater Elo (very likely) they'll gravitate to that bucket all on their own.

I'd say put it in the 'maybe some day' pile.

***edited to add***

Separate queues for teams.... how tough is that one? Seriously, like for a weekend in the test server. Just to see demand, response, performance. Teams of any size 2-12 plus the ability to pug in it.

It'd need a UI aspect, true. It certainly would screw up Elo but can Elo be ignored in the team queue? Then again it'd get the crap exploited out of it. Without Elo it'd be prone to stomps and wide swings, etc.

My concern, honestly, is that people want the *idea* of bigger teams, they just want bigger teams fit in seamlessly with the matchmaking we've already got. That's a much bigger thing to tackle.

That'd need some sort of synergy modifier for Elo, it'd need to track and keep Elo for not just every player but how every player plays with specific other players or the consistent presence of other people in their matches would skew their results. You might be able to just ham-fist it with giving premades an Elo multiplier equal to their % of the total size of their team and then dividing their k-factor for win/loss by how big a team they're in, preventing huge swings in the Elo on everyone in the premade.

End of the day I'd say you'd want to keep a separate Elo for the team queue (pugging or premade) and the solo queue.

Still though, lobbies and some sort of team size options that don't involve spending resources you're not making back would do so much for the game. It's not even about playing in a guild - it's about building connections with other players in the game, which in turn makes you care more about how they see your performance, which in turn makes you value your performance in the game more.

Edited by MischiefSC, 12 April 2014 - 04:55 PM.


#215 ApolloKaras

    Member

  • PipPipPipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 1,974 posts
  • LocationSeattle, Washington

Posted 12 April 2014 - 05:03 PM

My jaw dropped when I saw this thread, I didn't even know it existed until I saw on Reddit. Karl you're the best thing that's happened to PGI's communication since the founders package hit. Thanks for talking with the community here it does mean a lot.

#216 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 12 April 2014 - 05:46 PM

Mischief, the small change you describe may not move a premade between brackets, true. However, a 100 Elo points difference is already significant for team matching, so that "tiny" modifier becomes important.

And similar movement would happen when players switch between their strong and weak mechs within a class. Don't get me wrong, I love my Dragons to death, but nothing carries like a 3D.

Edited by Modo44, 12 April 2014 - 05:47 PM.


#217 p4r4g0n

    Member

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

Posted 12 April 2014 - 05:46 PM

Elo & Matchmaking - group Elo modifiers that apply based just on the number of people in the group seems like a pretty clumsy solution and adds another reason not to play in a group or at least limiting the group to the smallest possible. Do we really need another penalty for group play?

For simplicity's sake, wouldn't it be easier at least in the immediate time horizon to just have a solo drop Elo and a group drop Elo? This has been suggested before ... many times. It may not be as elegant as some of the suggestions involving individual / group tracking but it certainly seems a lot simpler to do as an interim measure if nothing else. Also, separate chassis Elo please in place of weight class Elo?

I also really like the suggestion to use the highest ratings of a percentage of the players in a group in place of the average group Elo rating. While it can be seen as somewhat punitive, it shouldn't matter in normal play and can be avoided by using free private matches if the group is being set up to help train a new player.

Edit: typo

Edited by p4r4g0n, 12 April 2014 - 06:56 PM.


#218 slide

    Member

  • PipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 1,768 posts
  • LocationKersbrook South Australia

Posted 12 April 2014 - 06:10 PM

With out getting into a debate on the pros/cons of burst fire on AC's.

Is there any technical limitation to having this in game?

I would have thought that the mechanics (lasers have about 60 ticks per beam) and effects were already mostly present within the engine. You can basically see the effects needed with UAC5 double shots, and chain fired AC2's.

Thanks for reading, love your input.

#219 Wintersdark

    Member

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

Posted 12 April 2014 - 06:46 PM

View PostShar Wolf, on 03 April 2014 - 03:43 PM, said:

Wait?!

So other people have that issue with other 1st person viewpoint games to?

I could never find much about that!

Yup, many, many people do. My wife can't be in the room if I'm playing other FPS type games, but MWO doesn't bother her at all.

#220 Alaskan Nobody

    Member

  • PipPipPipPipPipPipPipPipPipPipPip
  • The Determined
  • The Determined
  • 10,358 posts
  • LocationAlaska!

Posted 12 April 2014 - 06:49 PM

View PostWintersdark, on 12 April 2014 - 06:46 PM, said:

MWO doesn't bother her at all.

Well that is good news anyways. :)





11 user(s) are reading this topic

0 members, 11 guests, 0 anonymous users