1911 replies to this topic

[VanillaG]

Mawai, on 28 May 2014 - 06:21 AM, said:

<snip>
However, the goal is something that works BETTER ... not something that is perfect ... and since ELO by itself appears to have been demonstrated as insufficient ... then additional factors like mech or team effectiveness need to be factored in to match balancing ... and something other than weight class/tonnage might be a more accurate indicator of this.

The sticking point is that you have to quantify how much better that new algorithm is in comparison to just using ELO. The only way to do that is capture all of those stats at the time of match making so you can run them through some different models to figure out what if any changes need to be made.

It would suck to put dev resources into coding all of that up and realize that at the end that new algorithm is not that much better and you should have had those devs working on other features that provide higher impact. With the nature of how this game is being developed, the match maker is going to be re-written a couple of more times so trying to hit it out of the park at this stage is a little risky considering how many other things they can be working on.

[Cimarb]

VanillaG, on 28 May 2014 - 03:26 PM, said:

The sticking point is that you have to quantify how much better that new algorithm is in comparison to just using ELO. The only way to do that is capture all of those stats at the time of match making so you can run them through some different models to figure out what if any changes need to be made.

It would suck to put dev resources into coding all of that up and realize that at the end that new algorithm is not that much better and you should have had those devs working on other features that provide higher impact. With the nature of how this game is being developed, the match maker is going to be re-written a couple of more times so trying to hit it out of the park at this stage is a little risky considering how many other things they can be working on.

Neither Elo nor BV need to be calculated at the time of match making. They should both be calculated after the match, which I assume is when Elo is calculated/adjusted currently. If you have ever changed the data in a spreadsheet and hit "save", you know how long it takes to do the calculation.

[VanillaG]

Cimarb, on 28 May 2014 - 04:36 PM, said:

Neither Elo nor BV need to be calculated at the time of match making. They should both be calculated after the match, which I assume is when Elo is calculated/adjusted currently. If you have ever changed the data in a spreadsheet and hit "save", you know how long it takes to do the calculation.

It is not about the amount of time or computational power it takes to calculate that value, it is about the effort required to determine the value of the variables, which variables to use, and which equation to use. Then you quantify that the new algorithm is x% better than the old way. All of that work is non trivial and it needs to be prioritized among all of the other things that they can be working on.

I don't think anyone can say with a straight face that any new method of trying to make matches will be 50% better without running some simulations with live data. They are going to be re-writing the system from the ground up. In this first iteration just making it so you can run parallel algorithms at the same time in production would be better use of dev resources than taking a stab in the dark with a new brand new way of match making. Once you can conduct A/B testing in the live environment you can actually quantify which algorithm is better.

[p4r4g0n]

Cimarb, on 28 May 2014 - 04:36 PM, said:

Neither Elo nor BV need to be calculated at the time of match making. They should both be calculated after the match, which I assume is when Elo is calculated/adjusted currently. If you have ever changed the data in a spreadsheet and hit "save", you know how long it takes to do the calculation.

This would only apply if you are looking to determine how much to adjust the Elo rating of the players or match rewards bonuses based on the outcome. Don't forget that the primary intention of introducing Elo ratings was to help get fairer matches between players via pre-drop matching. 3/3/3/3 is just a further refinement to that end.

Having said that, the only alternative that I can see to 3/3/3/3, taking into consideration all the constraints about not limiting player choices too drastically as to what mechs they can play, is to revert to the equal matching of groups and weight class.

IMO the trouble we have with the current matchmaker boils down to the prioritisation of quick matches over fair matches. A range of 1400 compounded by the loosening of the weight class matching is the reason why we see so many ridiculous matchups. Or at least it was ..... I've lost track of exactly what criteria applies right now with all the changes, roll-backs, partial implementations,etc that have gone on.

While I sympathize with players whose Elo rating is at the extreme ends of the scale, realistically as the top / bottom 5% they should be looking to private matches if they want quick matches instead of jumping in the public queue.

Edited by p4r4g0n, 29 May 2014 - 12:34 AM.

[Tekadept]

This is a good read on different team Rating ideas. Made my head hurt a bit taking me back school math class but still a good read.

http://jmlr.org/pape...11a/weng11a.pdf

Edited by Tekadept, 28 May 2014 - 10:43 PM.

[Cimarb]

p4r4g0n, on 28 May 2014 - 10:27 PM, said:

This would only apply if you are looking to determine how much to adjust the Elo rating of the players or match rewards bonuses based on the outcome. Don't forget that the primary intention of introducing Elo ratings was to help get fairer matches between players via pre-drop matching. 3/3/3/3 is just a further refinement to that end.

Correct. Regardless of how you determine it, Elo is just a number popped into the matchmaker algorithm at the beginning of the match. It is then adjusted/calculated after the match. Any change to Elo would happen after, meaning there is no calculation (for Elo) before. The calculations happening before the match are completely different.

It's similar with mechs. When you change the loadout and hit save, it essentially calculates the value of that mech (currently it valuables solely on weight class, but ideally BV would be calculated to be more accurate). That value (class or BV) is then plugged into the matchmaker as a static number at the beginning of the match. All the work is done after the mech is saved, not when the match is formed.

[Karl Berg]

Cimarb, on 28 May 2014 - 11:05 AM, said:

In the VLOG that was just released, Russ mentioned that they are already trying out 4x4 (so up to four of any class), so I think they will have a flexible version pretty soon. I'm sure Karl will have a lot more to say on that point, if he is able.

Actually, I did enable 4's for a while.. And 5's, and 6's.. All the way up to 8's, before it simply became clear that the modifications that had been made to the current matchmaker were causing massive starvation within the request pool. Because of this, we had little choice but to revert back to the pre-patch matchmaker. The pre-patch matchmaker also 'supported' 3's, but enabling 3's caused it to hang, due to it's inability to handle production loads.

[Karl Berg]

Tekadept, on 27 May 2014 - 07:38 PM, said:

Hi Karl, just wondering if you have a stat on what is the average number of matches a player plays in a day?, ie 5 games a day, 10 games a day? And what is the biggest outlier, ie some poor soul doing 100 matches a day

Oh gosh.. D= I purposefully kept everything vague to within an order of magnitude so that this wouldn't happen. Sorry, I do have stats; but for whatever reason, contractual, data ownership issues, marketing, ??, we've been asked not to make online player numbers public.

[Tekadept]

Karl Berg, on 30 May 2014 - 01:34 AM, said:

Oh gosh.. D= I purposefully kept everything vague to within an order of magnitude so that this wouldn't happen. Sorry, I do have stats; but for whatever reason, contractual, data ownership issues, marketing, ??, we've been asked not to make online player numbers public.

Well you were happy to throw around other stats so can't blame a guy for trying.Guess I cant pull one over on you by using your previous "vague" number of logins you gave and then getting how many games the average person plays

Once I nearly accidentally let slip that.............. CRAP i nearly did it again..

How bout one single stat? what is the MOST games a person has played in a day recently? surely we cant use that to any ill effect

Edited by Tekadept, 30 May 2014 - 01:58 AM.

[Karl Berg]

BTW, for those of you asking to see my desk. Paul was sneaky, and put footage in vlog 4.

[Smitti]

I did see the TrackIR5 being tested in the Vlog, i believe we have you to thank for that Mr Berg!

[Clint Steel]

It seems to me that the best thing that will come from 4x3 is that teams will have similar mechs.

Couldn't the system just build teams that have same number of mechs in each weight class (ie. one team has 6 lights, the other team also gets 6 lights)? Even better, would be making sure each team has an equal number of ECM mechs, and other match tipping tech that might be known. If the point is closer matches that is.

[Hobo Dan]

Karl,
I asked this question in Suggestions, but I thought I’d throw it your way just in case. Is this remotely possible or likely?

Hobo Dan, on 30 May 2014 - 11:32 AM, said:

What if along with a spot assists, you received credit for damage done by teammate from your spotting? Give it the same weight as regular damage for match scoring purposes. Represent it as: direct damage(damage assist) on the match summary.
So if you’re running your spotter light and you only deal 152 direct damage, but the damage caused by your spotting (be it Tag, NARC or regular targeting) is around 450, your final stats look like this: 152 (450) and you are rewarded for doing 602 damage.
This is by no means a total fix for information warfare, but it seems like a relatively simple addition to help out those who do spot and encourage more to do it.

[Mawai]

Hi Karl,

I just wanted to put some feedback in a place where someone useful might see it

Paul mentioned the addition of a module that will break locks as soon as an opposing mech leaves line of sight.

A poster in the general forum raised the following issue:

"I think this will add more incentive to pop-tart and I for one would never pop-tart without this module."

I think he is absolutely right. It is hard enough already to fire at mechs that pop in and out of cover using jump jets and firing pinpoint direct damage weapons. Addition of this module would make them virtually immune to LRMs and immediate loss of targeting would make it difficult to tell whether they are moving to a new firing position or staying put. I am just wondering if this is an intended or unintended side effect of the design? Of course the module will benefit any mechs fleeing LRMs but I think it will help pop tarts more ...

Edited by Mawai, 30 May 2014 - 11:59 AM.

[Cimarb]

Hobo Dan, on 30 May 2014 - 11:46 AM, said:

Karl,
I asked this question in Suggestions, but I thought I’d throw it your way just in case. Is this remotely possible or likely?

Spotting assist does need to be a much higher bonus, I agree. Maybe it could be scored as half-points compared to actual damage done. Another issue is that it helps heavier mechs just as much as lighter mechs, but it would give an incentive for focusing your fire (since you would get a bonus based on the people you were helping focus with).

Regardless, we definitely need some more role-warfare incentives, as damage done is currently tops from what I can tell.

Mawai, on 30 May 2014 - 11:58 AM, said:

Hi Karl,

I just wanted to put some feedback in a place where someone useful might see it

Paul mentioned the addition of a module that will break locks as soon as an opposing mech leaves line of sight.

A poster in the general forum raised the following issue:

&quot;I think this will add more incentive to pop-tart and I for one would never pop-tart without this module.&quot;

I think he is absolutely right. It is hard enough already to fire at mechs that pop in and out of cover using jump jets and firing pinpoint direct damage weapons. Addition of this module would make them virtually immune to LRMs and immediate loss of targeting would make it difficult to tell whether they are moving to a new firing position or staying put. I am just wondering if this is an intended or unintended side effect of the design? Of course the module will benefit any mechs fleeing LRMs but I think it will help pop tarts more ...

I definitely agree with this issue. I'm hopeful that PGI has already thought about this, but it definitely seems like it will benefit jump snipers far more than anyone else.

[9erRed]

Greetings Mr. Berg,

Questioned earlier but it may not have been recognized as a question.

Q: Have you had any time to work on the coding for the artists to enable smaller 'rigid' body art elements?

You had posted that you were doing this 'on the side', and not ready to 'hand off' to the art department yet, but would that not require the artists to have to go back to all the older maps and complete all there checks for functionality and inclusion. Or would that just be for newer maps to start, and as time permitted, return to these past maps. (so looking forward at having something destructible in the 'Jungle' map.)

- Attack and Defend mode just screams for destructibility as part of the function of that mode.

Thanks,
9erRed

Edited by 9erRed, 30 May 2014 - 04:38 PM.

[East Indy]

Karl Berg, on 30 May 2014 - 01:22 AM, said:

Actually, I did enable 4's for a while.. And 5's, and 6's..

Were you able to glean anything from those looser constraints in spite of the matchmaking code's limitations? If so, what?

[Imperius]

I'm just sick of ELO "forcing me to lose" to see if I can overcome team potato.

It just translates to me getting utter mad and team-killing said potatoes. Nothing brings the troll out in me more that matchmaker telling me to carrry harder.

How I think ELO should work creating a close game of intense battle

• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro
  
• Pro Vs Pro

How ELO actually works create a game where I see 4 machine gun Jager 30,000 ammo think he can toe to toe with fresh Atlas DDC. While I pull my hair out and scream "Why the F*** is this potato on my team?"

• Pro vs Average
  
• Pro vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average
  
• Potato vs Average

Edited by Imperius, 31 May 2014 - 09:10 AM.

[Modo44]

I think it was said already, but why not hammer it in: Building teams instead of matches was not the best idea.

[SgtKinCaiD]

Dear Karl,

While you're working on the MM, are you going to review the ELO scoring, so that it is not based on W/L only ?

Thanks.