Jump to content

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


1911 replies to this topic

#1181 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 07 July 2014 - 04:18 PM

View PostCimarb, on 07 July 2014 - 01:13 PM, said:

Karl, Sasparilla Kid shared a screenshot and proved me wrong in my defense of the MM:

His 12-man went against at least two smaller groups, and both assault and heavy valves were released, giving the smaller groups a 2/2/4/4 spread compared to the 12-man 3/3/3/3 they are forced to use. I assume that is an extreme situation and MM just couldn't find them a match otherwise, but is that intentional? I am fine with valves if they still maintain the balance between teams, but a situation like this should not happen by what I have read.

(Edited to add SK's quote)


Hey Cimarb,

There is a 'timeout' safety feature for group queue, currently set at 4 minutes. If the system can't find a match for a group, even after 4 minutes waiting in the queue, it disables the strict weight class matching requirement. Matches are usually found quite quickly after this point.

We had tightened up Elo matching somewhat to reduce the number of very high Elo competitive teams dropping against lower-Elo premades, but the knock-on of tightening Elo was an increase in the number of groups hitting the 4 minute timeout. Those particular changes have now been rolled back.

This timeout value is set to infinite for the solo queue. Currently configured, it will never launch a solo queue match without strict weight class matching.

Sorry for the erratic behaviour everyone. At this point we're playing a delicate balancing game between Elo, 3's, weight class matching, and wait time, in an attempt to find the ideal settings on production for both the group and solo queues.

We had a good idea what the solo queue would look like prior to release, and so it was reasonably well tuned prior to launch. The grouping features are essentially new to the game, and I had little idea what frequency 5+ groups would queue, what their Elo's might be, what sort of weight classes they would prefer, etc; in addition to the group queue being significantly more difficult to balance by it's very nature.

#1182 Siriothrax

    Member

  • PipPipPipPipPip
  • The Defiant
  • The Defiant
  • 134 posts

Posted 07 July 2014 - 04:19 PM

Hey Karl, what are the chances of getting selectable ammo draw, eg choosing where ammo is used up from first? Right now, the order in which ammo is used is unchangeable, and probably very poorly known among the community. For some builds, the order in which they'd like to draw - using up ammo in locations that are going to be blown off first - is completely opposite to how the current system draws (torsos->arms->legs->head, with Left being used first for IS and Right being used first for Clans...which is kinda weird).

Edited by Siriothrax, 07 July 2014 - 04:21 PM.


#1183 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,573 posts
  • LocationParanoiaville

Posted 07 July 2014 - 04:31 PM

I wonder how well it would work if the matchmaker broke the Rule of Threes in increments. Say if by Time X you didn't have a group, it would allow 4 of whatever weight class is trending highest, then the two highest at Time Y, etc. - sort of a +/- 1 scaling for the Rule. I don't know if such a thing is already in place, or would require too much risk/effort to be prioritized, but it seems like it might yield smoother results if it could be implemented.

#1184 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 07 July 2014 - 04:34 PM

View PostSiriothrax, on 07 July 2014 - 04:19 PM, said:

Hey Karl, what are the chances of getting selectable ammo draw, eg choosing where ammo is used up from first? Right now, the order in which ammo is used is unchangeable, and probably very poorly known among the community. For some builds, the order in which they'd like to draw - using up ammo in locations that are going to be blown off first - is completely opposite to how the current system draws (torsos->arms->legs->head, with Left being used first for IS and Right being used first for Clans...which is kinda weird).


Hey Siri,

I'll forward this to gameplay and see if they have anything to say on it. I seem to recall hearing discussion on this a while back.

#1185 Goose

    Member

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

Posted 07 July 2014 - 04:50 PM

Sir, I'm complied to ask when was weight deemed more important then Elo?

Being out weighed in Skirmish is a real issue, yes, but in Assault, or more importantly, Conquest?

Doesn't this suggest something is out'a whack with the game modes?

#1186 Omid Kiarostami

    Member

  • Developer
  • Developer
  • 35 posts

Posted 07 July 2014 - 05:26 PM

View PostKarl Berg, on 07 July 2014 - 04:34 PM, said:


Hey Siri,

I'll forward this to gameplay and see if they have anything to say on it. I seem to recall hearing discussion on this a while back.


Hey Siri,

Whether we get selectable ammo draw is up to design and their priorities for getting stuff into the game. I can't really answer for them. Ammo draw ordering is fixed for all mechs and independent of faction. It's presently: Center Torso -> Right Torso -> Left Torso -> Right Arm -> Left Arm-> Left Leg -> Right Leg -> Head. No exceptions are intended, so if the the draw ordering is any different from that it's a bug.

Cheers,
Omid

#1187 GMan129

    Member

  • PipPipPipPipPip
  • Big Daddy
  • Big Daddy
  • 194 posts

Posted 07 July 2014 - 05:45 PM

View PostOmid Kiarostami, on 07 July 2014 - 05:26 PM, said:

if the the draw ordering is any different from that it's a bug.
Cheers,
Omid


It's reversed for at least the Mad Cat. I haven't done any specific testing on the other chassis, nor have I done testing on the arms, but the ammo gets taken out of the left torso before the right torso.

#1188 Cimarb

    Member

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

Posted 07 July 2014 - 07:47 PM

View PostGMan129, on 07 July 2014 - 05:45 PM, said:


It's reversed for at least the Mad Cat. I haven't done any specific testing on the other chassis, nor have I done testing on the arms, but the ammo gets taken out of the left torso before the right torso.

Is it maybe a cockpit display error (L/R swapped), or how are you testing it?

#1189 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,573 posts
  • LocationParanoiaville

Posted 07 July 2014 - 08:09 PM

View PostGMan129, on 07 July 2014 - 05:45 PM, said:


It's reversed for at least the Mad Cat. I haven't done any specific testing on the other chassis, nor have I done testing on the arms, but the ammo gets taken out of the left torso before the right torso.

Bah, I can't recall where I saw it, but some people are saying they've tested it out somehow.

#1190 Siriothrax

    Member

  • PipPipPipPipPip
  • The Defiant
  • The Defiant
  • 134 posts

Posted 08 July 2014 - 12:29 AM

View PostOmid Kiarostami, on 07 July 2014 - 05:26 PM, said:


Hey Siri,

Whether we get selectable ammo draw is up to design and their priorities for getting stuff into the game. I can't really answer for them. Ammo draw ordering is fixed for all mechs and independent of faction. It's presently: Center Torso -> Right Torso -> Left Torso -> Right Arm -> Left Arm-> Left Leg -> Right Leg -> Head. No exceptions are intended, so if the the draw ordering is any different from that it's a bug.

Cheers,
Omid


Hi Omid,

I understand that it's not a high priority, but it would definitely be something good to have. Consider the following: you have a fit with the main weapons in a side torso, and those weapons fill up most of the slots in that side torso so there's only one slot left. You need 4, maybe 5 tons of ammo for this weapon. You put one in the head, one in the same side torso - now where do you put the other two or three? Let's say you're a medium mech.

a - The arm for that side torso. Well, the torso is being drawn on first, and you're probably going to use that arm to shield, so if the arm is blown off you're left with nothing. You'd rather draw from the arm first, then the torso.
b - The other side torso. Well, again, if your weapons are in the RT, then you're going to be using the other one to shield...but it's going to draw on the RT first, so when you lose your shield side you're out of ammo. If your weapons are LT, you're good to go, but *most* mechs are mostly RT mounted.
c - The legs. A decent choice, but one of the primary causes of death for mediums is legging. Unfortunately, the torso is drawn first, so if you get legged, there's a decent chance that you're going boom from the additional crit rolls. You would want to draw on this ammo first, so that when you're legged, there is less or zero ammo to be crit.

These are just some of the main examples of the drawbacks of the currently-inflexible system. I do hope that you guys are able to find a solution. :( In the meantime, there are a couple fits I have that just straight up do not work with the current system.

Edited by Siriothrax, 08 July 2014 - 12:32 AM.


#1191 Mawai

    Member

  • PipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 3,495 posts

Posted 08 July 2014 - 07:50 AM

Hi Karl,

I was wondering if you could say something about weapon convergence or rather the lack of it.

At present all weapon fire appears to be traced from the weapon mount location to the intersection point of the reticle with the terrain or target mech (whatever it is aiming at). This is instantaneous so if the reticle moves off a mech all weapons can end up aiming at distant terrain (firing nearly parallel) and miss the target even if the reticle is off by one pixel. Lasers can be particularly affected as they go from hitting to missing even with small changes in aim.

However, it seems that convergence was intended to be implemented (it may have been at one point?) since there is a tier 2 mech efficiency to improve weapon convergence time that currently does nothing as far as I know.

Convergence has several game play effects in reducing the effectiveness of pinpoint damage and certain tactics that go along with it. Targeting has to be held for a certain length of time before a target is "locked" and the firing weapons converge on the reticle. This increases the skill required to effectively deal damage and means that simply holding the reticle on a target for the instant that you fire is insufficient to deliver all of the damage to a single component.

I was wondering what the status of weapon convergence is? Does it put too great a load on the server? Are there other technical issues preventing implementation? It is clearly something that has been considered since closed beta ... so I was wondering if you'd be able to provide an update?

P.S. You could probably implement convergence fairly efficiently by defining the target point for each weapon firing randomly within a circle around the aim point ... perpendicular to the line of fire. The radius of the circle would decrease with the time the reticle stays focused within a bracket at a given range. The hardest part might be developing an efficient algorithm to track the size of the aiming circle for each mech on the field in real time.

Edited by Mawai, 08 July 2014 - 07:57 AM.


#1192 Shlkt

    Member

  • PipPipPipPipPipPip
  • Ace Of Spades
  • Ace Of Spades
  • 319 posts

Posted 08 July 2014 - 10:35 AM

Karl,

I've read several anecdotes regarding the relationship between match quality and time of day. With the new matchmaker data would it be possible to produce a heat map of Elo variance within a team vs. time of day? It would be interesting to see just how much the time of day affects the potential skill diversity in a match.

#1193 Wintersdark

    Member

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

Posted 08 July 2014 - 11:38 AM

View PostShlkt, on 08 July 2014 - 10:35 AM, said:

Karl,

I've read several anecdotes regarding the relationship between match quality and time of day. With the new matchmaker data would it be possible to produce a heat map of Elo variance within a team vs. time of day? It would be interesting to see just how much the time of day affects the potential skill diversity in a match.
both in terms of team vs. Team and in terms of overall player Elo delta - I've long been very curious what the average delta is within a team. (Re: stories about extremely goof pilots being matched with poor pilots to balance team Elo, particularly against an average-Elo team.

#1194 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 09 July 2014 - 03:32 AM

One thing that maybe was not addressed well, and which may hurt the matchmaker unduly. In chess, you can start a player's Elo in the middle because the rules are obvious -- the early learning curve is flat, and one person's Elo does not affect other matches directly. MWO's early learning curve is extremely steep, on the other hand, which means that new players are generally never in the ~1300 range they start off. Maybe this value should be lowered (considerably) to prevent obviously recurring instances of complete noobies getting pitted against veterans.

Incorrect matching may not be a problem for one player -- it will converge, no biggie. However, it biases the matchmaker with essentially all green players having incorrect Elo values. If you match rookies only vs other rookies, you give them undue credit on match 26. If you match them with anyone, you guarantee very bad matching on a potentially large player pool from the very beginning. Since you always have new players, that is essentially constant bias.

Relatively high starting Elo may mean constantly overvalued players which may result in mismatches despite the values looking good. Say you build one team of relatively bad veteran players and another team with completely green people, both with an average 1300 Elo. Now, the veterans may not be stellar, but they at least bring upgraded mechs, and basic game+map knowledge, potentially even decent aim. Specifically because MWO is so hard, being average already requires considerable skill. What is OK in chess with the 1:1 mechanic and clear, simple rules, may be consistently hurting the MWO matchmaking.

Edited by Modo44, 09 July 2014 - 03:36 AM.


#1195 Noesis

    Member

  • PipPipPipPipPipPipPipPipPip
  • Legendary Founder
  • Legendary Founder
  • 4,436 posts
  • LocationIn the Lab

Posted 09 July 2014 - 06:12 AM

It would make sense in theory, but could be capatalised on at tournament time or as veteran players starting with alternative characters. The later of the two less of a potential issue I guess when transferance of infrastruture is not apparent. But c-bill gifting has been identified as contributory from players to units (even just as tax) whenever this comes into effect, so allowing newer accounts on a free server to have better performance due to being sympathetic to ELO could allow for more easier farming with alts by the more skilled players. This more of an attraction with the application of cadet bonuses.

Thus if the above becomes prevelant as an activitiy you could then still end up with the issue of skilled players having alt accounts still playing with newer pilots as a result since they (will for a time) have a more average ELO. The question being would you want to extend the attraction of this capability?

Overall the law of averages with the culmination of a full team will try to balance ELO settings in theory and place a player into an area relevant to their skill set and having an expectation you may still be on a team with more or less capable players.

The only really safe way to safeguard newer players would be to have a seperate queue, assuming player populations and technical provisions could support this sensibly. This at least might afford a time to at least shake off the "greenhorn" issues prior to joining the public queues. But this then in theory also then allowing for an area for alts of existing players to potentially capitalise on new players.

--

As a more constructive thought, I hope, the learning cliff for newer players can also help to promote the idea of grouping with others and capatalising on various training and supportive techniques through awareness from others, be it IG tutorials, new player forums on MWO and/or the attraction of joining player units that provide both community and training techniques as supportive measures.

Edited by Noesis, 09 July 2014 - 06:41 AM.


#1196 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 09 July 2014 - 08:48 PM

Gifting between players is not an option in MWO. High-Elo players already make a killing if they decide to make alt accounts. Given how fast Elo converges when you carry hard (BTDT), we are talking about under 10 matches of difference with a lower starting value. Meanwhile, the vast majority of players seems to start off with extremely inflated Elo, acting as a constant stream of bad matchmaker predictions.

#1197 Void Angel

    Member

  • PipPipPipPipPipPipPipPipPipPip
  • The Marauder
  • The Marauder
  • 6,573 posts
  • LocationParanoiaville

Posted 09 July 2014 - 09:09 PM

Starting players at a much lower level would have an inflationary effect on the rating system if it were set too low. I'd recommend doing a statistical analysis of new players (discarding a certain proportion of outliers, or ideally duplicate accounts if they can be identified) before adjusting the entry Elo level.

If you find that the starting Elo is too high, Modo's suggestion seems like an excellent idea.

PS: Elo is not an acronym. =P

Edited by Void Angel, 09 July 2014 - 09:09 PM.


#1198 Hex Pallett

    Member

  • PipPipPipPipPipPipPipPipPip
  • Urban Commando
  • Urban Commando
  • 2,009 posts
  • Twitter: Link
  • LocationHomeless, in the streets of Solaris 7

Posted 09 July 2014 - 09:15 PM

How about a "noob chamber"? A new player queue specifically for newcomers? Most mass-market trashy F2PFPS games have that, and Smite, the last MOBA I played, have similar practice queues as well.

You do risk having high-skill player using alt accounts trolling new players, but I think most of us are too busy sticking to and investing on a single account....

#1199 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 10 July 2014 - 12:31 AM

The biasing happens whenever players enter the ranked queue with a preset Elo value. Separating them before they do will not change the main issue.

View PostVoid Angel, on 09 July 2014 - 09:09 PM, said:

Starting players at a much lower level would have an inflationary effect on the rating system if it were set too low. I'd recommend doing a statistical analysis of new players (discarding a certain proportion of outliers, or ideally duplicate accounts if they can be identified) before adjusting the entry Elo level.

Yeah, hence all the "maybe" in my post. Just remember that when looking at current average Elo values -- even for people with thousands of matches -- you are looking at results skewed by constantly incoming 1300 Elo rookies.

#1200 Cimarb

    Member

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

Posted 10 July 2014 - 06:03 AM

View PostHelmstif, on 09 July 2014 - 09:15 PM, said:

How about a "noob chamber"? A new player queue specifically for newcomers? Most mass-market trashy F2PFPS games have that, and Smite, the last MOBA I played, have similar practice queues as well.

You do risk having high-skill player using alt accounts trolling new players, but I think most of us are too busy sticking to and investing on a single account....

There is already a cadet queue that does that. First 25 matches. It is not enough, though, as 25 matches is not a whole lot.

View PostModo44, on 10 July 2014 - 12:31 AM, said:

The biasing happens whenever players enter the ranked queue with a preset Elo value. Separating them before they do will not change the main issue.

Yeah, hence all the "maybe" in my post. Just remember that when looking at current average Elo values -- even for people with thousands of matches -- you are looking at results skewed by constantly incoming 1300 Elo rookies.

Elo is horrible, and mostly for that very reason. I realize I have an affect on my match, and I bust my behind making sure I am one of the top contributors to every match, but when you are dropping solo with 11 random people, the amount of "random" you have to overcome is immense. I highly, HIGHLY dislike being measured by that randomness (aka W/L the only determining factor for Elo), and the arguments that say "it will average out and give you the right number" are horrible.

If you average ANYTHING for long enough, it will be average - it is the law of averages, lol.





1 user(s) are reading this topic

0 members, 1 guests, 0 anonymous users