Jump to content

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


1911 replies to this topic

#1141 Cimarb

    Member

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

Posted 03 July 2014 - 05:24 AM

View PostModo44, on 02 July 2014 - 10:11 PM, said:

Now that the matchmaker makes sense, there are two obvious improvements:

Track group and solo Elo separately. You keep talking about adding a set modifier, but that has obvious issues. You can not predict who sucks the same (or even more) when grouped up. Pushing group Elo manually would introduce new problems, specifically: unduly punishing bad players by asking them to carry unrealistically hard. Tracking group Elo would automatically adjust for every player's group play (in)ability.

Track chassis Elo separately. Seeding issues were mentioned for this. One solution would be using MWO-specific battle values (i.e. averaged Elo values for each chassis -- already tracked) to help with initial seeding and re-seeding of chassis Elo. Again, no manual adjustments, just working with heuristics based on Elo.

There are many issues with both of those, though.

Group/solo Elo: not only do you have a LOT of variables to worry about, you also have to consider that some people group differently with different groups. For instance, when I am grouped with my normal lance, I am going to do quite well, but if it is not a practice night I often find myself grouping with random people from my unit that I generally have never played with before. Would I have the same group Elo? What if I did good with the three people I initially group up with, and then a fourth player comes in and joins the group and we start doing horribly? Since we are doing horribly, one of the initial 3 drops out, so we are back down to a group of four and start really clicking. Obviously, the issue was between the fourth player and the one that dropped out, or was it? Maybe it just took the fourth player a little while to shake the rust off individually, and now that he has, we would have done well even if the other person had stayed. WAY too many things going on...

Chassis Elo: much like Roland's idea of "value by popularity" in another thread, I am not a fan of doing Elo based upon mech chassis, since there are way too many things deciding how well you do within a chassis, especially loadout. The Clan Omnipod system makes that even more difficult. I think a much better way would be to have an adjustment to your Elo (or just base "class" balancing upon this number instead of straight weight), based upon the combined Battle Value of all components equipped on your mech. This would give a very easily calculated, but highly accurate, number to balance with.

#1142 DEMAX51

    Member

  • PipPipPipPipPipPipPipPipPip
  • Elite Founder
  • 2,269 posts
  • LocationThe cockpit of my Jenner

Posted 03 July 2014 - 09:04 AM

Hey Karl, I have an interesting question about the in-game physics.

Since the advent of fall damage, and the f/s speed indicator on the HUD, people have realized that either 'Mechs are only about 1/3rd the size they should be (i.e. an Atlas is only 17 feet tall, instead of 17 meters), or the force of gravity on 'Mechs is 3 times that of actual gravity.

Can you tell us which it is?

Edited by DEMAX51, 03 July 2014 - 09:05 AM.


#1143 Goose

    Member

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

Posted 03 July 2014 - 09:48 AM

The gravity is way up: Remember the thing about lights couldn't run downhill on Mordor? Always catching "big air" when they tried to do so?

Yeah: Clan equipment, and the gunnery skill that came with it, obsoleted Lights in table-top; The "downhill thing" would coffin-nail them the same way the missing water movement penalty would also, without the better guns …

#1144 Deathlike

    Member

  • PipPipPipPipPipPipPipPipPipPipPipPipPipPip
  • Littlest Helper
  • Littlest Helper
  • 29,240 posts
  • Location#NOToTaterBalance #BadBalanceOverlordIsBad

Posted 03 July 2014 - 05:01 PM

Karl... is there going to be a Command Chair post about the current MM? A writeup of what now it does would be helpful (instead of just conjecture through experience).

Also, what is the current max # of premades on one team? Is it 3 or 4?

The answers to these questions would be greatly appreciated.

#1145 slide

    Member

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

Posted 03 July 2014 - 06:41 PM

Hi Karl,

I would like to thankyou on behalf of all 100+ active members of ARMD for helping to bring back groups larger than 4.

I know it was your skills and knowledge that would have made it possible. Even after only 1 night we are seeing a rejuvenation and renewed vigour from new and old members alike. Now we just need to dust of some of those rusty old skills from closed beta.

Just as an anecdote: I dropped with 7 others last night with 2M, 3H, 3A and was on the verge of getting people to change out of an assault and a heavy when we decided to drop and see what MM came up with to fill out the team. Australian peak time is probably one of the hardest for MM to fill in spots and I thought it would fail when having to find 3 lights and a medium. To my surprise after only about a 2 minute wait it launched us with 2 lights and 2 mediums, which I thought was an excellent outcome considering the likely difficulties.

Thanks again.

#1146 TheCaptainJZ

    Member

  • PipPipPipPipPipPipPipPipPip
  • The CyberKnight
  • The CyberKnight
  • 3,658 posts
  • LocationUnited States

Posted 03 July 2014 - 09:44 PM

View PostDeathlike, on 03 July 2014 - 05:01 PM, said:

Karl... is there going to be a Command Chair post about the current MM? A writeup of what now it does would be helpful (instead of just conjecture through experience).

Also, what is the current max # of premades on one team? Is it 3 or 4?

The answers to these questions would be greatly appreciated.

They do plan on doing a write up soon. The max number of premades in the solo queue is 1 on each team if and only if there is one on the enemy team. The group queue I assume has no set amount. Whatever makes up to 12.

Edited by TheCaptainJZ, 03 July 2014 - 09:45 PM.


#1147 Nothing Whatsoever

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • Bad Company
  • 3,655 posts
  • LocationNowhere

Posted 03 July 2014 - 11:38 PM

I've been dumb in the past in asking questions here, but...(lets see)...


@Karl... thank you dude!

I need to say matches have felt very fun; as long as my internet has held up!

Really great work with the MM.

Kudos!

#1148 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 03 July 2014 - 11:46 PM

View PostCimarb, on 03 July 2014 - 05:24 AM, said:

based upon the combined Battle Value of all components equipped on your mech. This would give a very easily calculated, but highly accurate, number to balance with.

The issue with that idea is prominent in other games which use battle value (WoT, WT, for example). You calculate values based on some game balance, but those values also influence game balance, so you get a new game balance, so you recalculate values based on the changed balance, so you get a new... It never, ever ends unless you make all player choices equal (i.e. have all mechs essentially reskinned versions of the same thing). This is unavoidable because players always adjust. You end up with dev time constantly devoted to "fixes" that become moot very quickly.

Separately, you are punishing players who are bad in high BV equipment, and you are giving handouts to people who happen to be great in low BV equipment.

All in all, BV is a terrible system when better tools are available. Remember that Elo tracking is not really viable in the environment BV was invented for -- casual TT games. Automatical adjustments based on individual skill will always trump a manually controlled system.

Edited by Modo44, 03 July 2014 - 11:51 PM.


#1149 Cimarb

    Member

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

Posted 04 July 2014 - 05:41 AM

View PostTheCaptainJZ, on 03 July 2014 - 09:44 PM, said:

They do plan on doing a write up soon. The max number of premades in the solo queue is 1 on each team if and only if there is one on the enemy team. The group queue I assume has no set amount. Whatever makes up to 12.

Group queue tries to balance group size as well.

View PostModo44, on 03 July 2014 - 11:46 PM, said:

The issue with that idea is prominent in other games which use battle value (WoT, WT, for example). You calculate values based on some game balance, but those values also influence game balance, so you get a new game balance, so you recalculate values based on the changed balance, so you get a new... It never, ever ends unless you make all player choices equal (i.e. have all mechs essentially reskinned versions of the same thing). This is unavoidable because players always adjust. You end up with dev time constantly devoted to "fixes" that become moot very quickly.

Separately, you are punishing players who are bad in high BV equipment, and you are giving handouts to people who happen to be great in low BV equipment.

All in all, BV is a terrible system when better tools are available. Remember that Elo tracking is not really viable in the environment BV was invented for -- casual TT games. Automatical adjustments based on individual skill will always trump a manually controlled system.

You calculate the values based upon the stats of the weapon. So a certain damage is worth a set amount, regardless of weapon, cooldown is worth a set modifier, etc. You are not balancing the weapons values against each other; you are assigning a value automatically based on the combined stats of a particular weapon. Yes, some dev work would be involved in initially setting the values, and like everything in the game those values could need to be adjusted periodically, but very little of it is manual.

#1150 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 04 July 2014 - 06:56 AM

Missing the point, again. How much equipment is worth on average depends on how the players use it. Equipment usage depends on how you value the equipment. It is perfectly circular, and can not -- by definition -- ever be calculated reliably and defined as a constant.

And, again, it completely ignores how good or bad any individual player is with specific equipment. Battle value could be great for seeding Elo as it provides a "currently average capabilities" look for any specific mech or build. But it will just not work as anything more. It has been tried, and it fails spectacularly in every single online game.

Both Elo and BV try to do the same thing, and they break each other when combined. Elo is looking to find everyone's specific skill with each tracked value (weight class at the moment), but it does not care what is average. BV is tracking the average value of equipment, but -- by definition -- can not adjust for any one player.

Edited by Modo44, 04 July 2014 - 07:02 AM.


#1151 Cimarb

    Member

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

Posted 04 July 2014 - 09:41 AM

View PostModo44, on 04 July 2014 - 06:56 AM, said:

Missing the point, again. How much equipment is worth on average depends on how the players use it. Equipment usage depends on how you value the equipment. It is perfectly circular, and can not -- by definition -- ever be calculated reliably and defined as a constant.

And, again, it completely ignores how good or bad any individual player is with specific equipment. Battle value could be great for seeding Elo as it provides a "currently average capabilities" look for any specific mech or build. But it will just not work as anything more. It has been tried, and it fails spectacularly in every single online game.

Both Elo and BV try to do the same thing, and they break each other when combined. Elo is looking to find everyone's specific skill with each tracked value (weight class at the moment), but it does not care what is average. BV is tracking the average value of equipment, but -- by definition -- can not adjust for any one player.

No, you are missing the point, actually. BV is independent of skill. It does not matter how YOUR skill affects the equipment, it is solely to determine the VALUE of the equipment. The damage, cooldown, heat and all other factors for a Medium Laser do not change because you are using it - YOU have nothing to do with the value of that item. You are trying to say that a dollar bill is worth more in your hand than mine, and that is just not the case.

Now, that dollar bill (or Medium Laser), can be used much more EFFECTIVELY by you, because you are a better shopper, but the value of that item does not change regardless. It is still $1 (or still a ML), regardless of how well you use it. The VALUE has not changed.

So, $1 is the BV of a ML. Whether you spend it on "5 20-cent items" by doing 1 point of damage on 5 different components, or "1 100-cent item" by doing 5 points of damage on 1 component, is a matter of your SKILL, ala Elo. Either way, the BV has not changed, just your application of that value.

#1152 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 04 July 2014 - 09:45 AM

View PostCimarb, on 04 July 2014 - 09:41 AM, said:

No, you are missing the point, actually. BV is independent of skill. It does not matter how YOUR skill affects the equipment, it is solely to determine the VALUE of the equipment.

Yes. Values average, and by adding them to the equation, you break Elo. You can then allow for Elo to re-converge, and keep BV static, which will mean "new" Elo with literally no improvement to the system. I already explained what happens if you try to re-adjust BV as game balance changes. All in all, a perfectly useless effort.

#1153 Cimarb

    Member

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

Posted 04 July 2014 - 11:19 AM

View PostModo44, on 04 July 2014 - 09:45 AM, said:

Yes. Values average, and by adding them to the equation, you break Elo. You can then allow for Elo to re-converge, and keep BV static, which will mean "new" Elo with literally no improvement to the system. I already explained what happens if you try to re-adjust BV as game balance changes. All in all, a perfectly useless effort.

Kind of like trying to balance by weight classes AND Elo, huh? All BV is is a better way to valuate mechs, just like they currently do with weight class. BV just happens to be much more accurate and takes loadouts into account. It wouldn't "break" anything, any more than it already is, of course.

#1154 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 05 July 2014 - 11:15 AM

View PostDEMAX51, on 03 July 2014 - 09:04 AM, said:

Hey Karl, I have an interesting question about the in-game physics.

Since the advent of fall damage, and the f/s speed indicator on the HUD, people have realized that either 'Mechs are only about 1/3rd the size they should be (i.e. an Atlas is only 17 feet tall, instead of 17 meters), or the force of gravity on 'Mechs is 3 times that of actual gravity.

Can you tell us which it is?


Hey there DEMAX51, as far as I've been told, mechs are scaled to 1 in-game unit == 1 meter. I'll have to double check though, since I haven't been very involved with the art / gameplay pipeline.

#1155 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 05 July 2014 - 11:19 AM

View PostShinVector, on 30 June 2014 - 06:31 PM, said:


You know Karl... That almost sound like the same problem that we have with lasers HSR and moving targets. Even though we see the hit, we always get poor damage and inaccurate damage.

It feels like HSR is not working at all of lasers most of time. If the server is using 'unrewound positions' to calculate laser damage.... That would explain we feel the need to lag lead shoot to get better damage with lasers.

I hope Neema is able to investigate (or already investigating) this possible issue.


Lasers do have their own issues, and we have located some unrelated potential causes for reduced laser accuracy. Lasers trace fire and terminate at the hit point however, so they don't share the same distance check mechanic for spread damage that SRM's have.

#1156 Cimarb

    Member

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

Posted 05 July 2014 - 11:20 AM

Karl, could you look at this MM thread when you get a chance? There are some interesting stats, but I would love to get your input on it because it is obviously a very small sample size.

#1157 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 05 July 2014 - 11:27 AM

View PostCimarb, on 30 June 2014 - 06:33 PM, said:

Well, to be totally honest, I knew it was slightly different and I was hoping to just evolve the question to get the answer I wanted. I fail at doing that quite horribly, so it didn't work :lol:

I know you have been asked the question A LOT, but that is because we have never received an answer! If we get an answer, you won't have to field the question again ^_^

So............... is a 4-man group considered a single "drop", or is it 4 "drops"?


For matchmaking concerns I care about total player counts within each queue.

I *do* know what the group / player mixture post new matchmaker launch; both in terms of requests and in terms of normalized player percentages. The command chair numbers don't line up with either of these number sets.

#1158 Karl Berg

    Technical Director

  • 497 posts
  • LocationVancouver

Posted 05 July 2014 - 11:34 AM

View PostCimarb, on 05 July 2014 - 11:20 AM, said:

Karl, could you look at this MM thread when you get a chance? There are some interesting stats, but I would love to get your input on it because it is obviously a very small sample size.


Hey Cimarb, unfortunately those variance numbers are not true statistical variance values. Out of lazyness I was simply calculating highest Elo on team - lowest Elo on team, rather than a true normalized sum of differences from mean.

Those screenshots were also taken within the first hour of launch of the new matchmaker. We've gone through extensive tuning passes since, both to bring wait times vs percentage 3's games in-line with stakeholder expectations, and to significantly tighten up Elo matching as well.

#1159 Modo44

    Member

  • PipPipPipPipPipPipPipPipPip
  • Bad Company
  • 3,559 posts

Posted 05 July 2014 - 11:37 AM

View PostCimarb, on 04 July 2014 - 11:19 AM, said:

Kind of like trying to balance by weight classes AND Elo, huh?

Yeah, only breaking things further by adding an elaborate system to tune on top of the existing ones.

#1160 Cimarb

    Member

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

Posted 05 July 2014 - 11:38 AM

View PostKarl Berg, on 05 July 2014 - 11:34 AM, said:


Hey Cimarb, unfortunately those variance numbers are not true statistical variance values. Out of lazyness I was simply calculating highest Elo on team - lowest Elo on team, rather than a true normalized sum of differences from mean.

Those screenshots were also taken within the first hour of launch of the new matchmaker. We've gone through extensive tuning passes since, both to bring wait times vs percentage 3's games in-line with stakeholder expectations, and to significantly tighten up Elo matching as well.

That is awesome to hear! I will share this on my thread so people know there too.

Any chance we can get an updated sample, maybe? ^_^





2 user(s) are reading this topic

0 members, 2 guests, 0 anonymous users