140 replies to this topic

[KingCobra]

GreenJulius ( I don't think the CryEngine has built in matchmaker options. The built in AI is for character AI, not match or server based logic. The matchmaker in use by MWO is almost surely a construct of their own. A monster without compare)

First you have to understand green the MM AI can be set to recognize players as characters it can control even the damage output of player weapons or reduce player damage because it controls everything server side. The serve AI is set to control the outcome of matches by matching player high or low ELO or even the damage ratios of player vs player.

That is why the server AI and ELO MM needs to be shut down and the MM run like in past MechWarrior PC games it was skill Vs skill with no server AI interaction.You could match players manually by tonnage or weight classes.

I did not want to believe even my damage output was being manipulated by the server AI but when I tested it over 100 battles on multiple new accounts I started to notice a pattern of high wins/kills until I reached 60-40% win/losses then no matter what I did I would start to lose matches and get killed repeatedly until my win/loss percentage was back to 50/50%.

Edited by KingCobra, 02 September 2014 - 06:47 PM.

[That Dawg]

El Bandito, on 02 September 2014 - 01:23 AM, said:

Especially since groups are now gone from the solo-q.

I posted a screenshot of six or so doing a timed drop. Got deleted. I see it once a night or so, some "unit" will sync drop in, probably because they aren't that good, but its annoying.

I still think MM deals your win, or your loss when you click "ready"

And dealing weight would be the quickest easiest way to balance.
Launch with one side who has all assaults and they have 8, the otherside composed of mediums and lights, gets a full 12, or variations of that.
Tonnage, and launch

[KingCobra]

ThatDaAwg (I posted a screenshot of six or so doing a timed drop. Got deleted. I see it once a night or so, some "unit" will sync drop in, probably because they aren't that good, but its annoying.)

These same groups have been sync dropping in solo queue for ever they cannot play in the group queue because they have no skill and prefer to Seal Club low or new players.In truth these groups of players must have very small Epeens to have to challenge players with so low of ELO rating there just punks that truly need to be banned.

PGI knew about the sync drop problem long before even the split of solo and group MM queues.But they still allow it because these players spent money.

[Bhael Fire]

Each player really should have two sets of Elo scores; one for solo playing and one for when they in a group. This would help get more accurate matchings.

However, they really messed up by basing their matchmaker on the Elo rating system in the first place — it's a system that works much better in a persistent 1 vs 1 environment or one where the team members remain relatively the same on regular basis.

Removing groups from the solo queue was huge step in the right direction, but there's much more work to be done before MM will be able to perform well consistently.

[Duke Nedo]

Well, all I can say after reading through is that if the ELO actually does force losses by making you carry dead weight, then it means that the MM deliberately creates uneven games. That would be the most horrific game developer choice ever. Forever ever.

My concluson is that we really need experience brackets instead. Skill in a team game where the same person can drop in a different mech with a different loadout every time is too complex to rate correctly. I say 3 brackets are enough, should give large enough player pools, then balance the teams on tonnage only. I dont think ecm should be in the MM, but I do think that ecm needs a full revamp to make it less dominant and open up for information warfare.

Forced losses, if a reality, needs to go. It's like telling the game to give the customer a bad experience that he cannot affect, that's just a complete gamebreaker...

[Thrudvangar]

KingCobra, on 02 September 2014 - 07:31 AM, said:

I would suggest to PGI just turn of the MM AI and let players just be placed in matches with no ELO except new players for there first 50 games just match by tonnages only let the player play on skill alone.

thats what i'm saying! Drop those bullshit like MM and ELO. nobody needs it, it always fails. just bring the ppl randomly into a match and let them play.

this elo **** is like the arena rating in WoW.. BUT MWO isn't WoW, we dont want such an arena ****. bring 'em all together and see what happens. ppl will have fun again.... perhaps....

[Zolaz]

No one is going to be happy when PGI explains their Match Maker. Well, except Blood Wolf he loves PGI like a fat kid loves cake.

[Duke Nedo]

Didn't we read that they now can tweak the MM seamlessly without patching?

[MrMadguy]

In theory forced 50% winrate MMR system isn't so bad. It should try to offer you enemies with skill level, close to your own, which leads to "natural" fair 50% chance to win and balanced matches. But only when properly implemented. This constant wide swaying or on the contrary - constant getting stuck in one of extreme states, are caused by poor distribution of players on two opposite teams. Team with more even skill levels gets big advantage.

[Sandpit]

Here's the thing (that most people forget or don't understand regarding Elo and MM)

You go on a nice winning streak. You eventually begin facing opponents of higher skill level. It's not the MM or Elo trying "even out" your W/L records, it's simply your Elo on a steady increase until eventually you're playing against opponents "better" than you skill wise.

it's like equating a college football player being drafted to the pros and realizing that the speed of the game has increased dramatically. It's not that the player has gotten "bad", it's that he is now playing with others at the same or higher skill level than he is at
on the entire team
as opposed to facing off against 1-2 superstars on the opposing team.

Same principle here. The MM keeps lining up opponents and nobody complains about the 10 game win streak they just came off of before going on that 12 game losing streak.

now that's not to say the MM and Elo couldn't use some work. There are still the issues of new players no longer being considered "new" after only 25 matches.
Elo being based on the team's effort as opposed to individual effort (which is why you see players with ridiculous KDR but mediocre W/L stats).
Elo is based on individual skill, it's not meant to be a "team" parameter. When you start taking averages of Elo, you've borked the whole system.
Player A = 200 Elo
Player B = 800 Elo
Average Elo = 500
That means Player B is probably going to do quite well and those players are getting the 600+ damage and multiple kills. Meanwhile, Player B is playing well above their Elo and the ones doing sub 200 damage and getting 0 kills.
Now toss in the ability to game Elo because it's set up by weight class instead of individual mech and players have the ability to "throw it off" even more.

I know a few players who keep their Locusts around just so they can have a "crappy" mech to pull out and keep their light Elo down. They don't care about that mech's stats so going in and doing poorly doesn't bother them. This drops their Elo for the entire weight class so when they pull out that optimized Jenner, they go in (at the lower Elo rating) and clean house. Then when they start losing, they switch back to the Locust.

These are the problems with how PGI has implemented Elo, not with the Elo system itself.

Duke Nedo, on 03 September 2014 - 04:03 AM, said:

Didn't we read that they now can tweak the MM seamlessly without patching?

yes they stated that

[Duke Nedo]

Sandpit, I would have loved it if it worked like that because if it did I would get good opponents AND good teammates. The problem is that when the loss steaks kick in, the opponents are much, much better than my teammates. I mean really much better. The screenshot I posted before had one trial mech and one guy asking if you can lock with SRMs in my lance. That really should not happen because my elo went up....

...so I just think something is really buggy with the MM, or they added some layer of systematic control that went bananas. Would be nice to get a statement if PGI thinks it is working as intended.

[Sandpit]

Duke Nedo, on 03 September 2014 - 09:21 AM, said:

Sandpit, I would have loved it if it worked like that because if it did I would get good opponents AND good teammates. The problem is that when the loss steaks kick in, the opponents are much, much better than my teammates. I mean really much better. The screenshot I posted before had one trial mech and one guy asking if you can lock with SRMs in my lance. That really should not happen because my elo went up....

...so I just think something is really buggy with the MM, or they added some layer of systematic control that went bananas. Would be nice to get a statement if PGI thinks it is working as intended.

Which is why I showed the averaging example. That's how a player winds up outclassed

The combined team Elo might be 2000
but if you have 6 players at 18000
and 6 players at 6000, that averages to 2000 Elo, the problem is the MM system think it's an "even" match when the other side has 12 players at 2000 Elo

Both of those teams have an Elo of 12000 in total but one team is severely hampered by having half of the team composed of players so far below the other team's Elo. That's the problem with how PGI has implemented Elo.

You don't have a true Elo per mech, you have an average Elo per weight class
Your team doesn't have a true Elo, it has an average based on all players.

Then you throw in the release valves that expand the acceptable Elo ranges and you have what we have now. It's not Elo, it's how Elo has been implemented here. Until that changes the MM is not going to perform any differently. It can't, it can only perform based on parameters set down by PGI.

[Willard Phule]

Sandpit, on 03 September 2014 - 07:52 AM, said:

Here's the thing (that most people forget or don't understand regarding Elo and MM)

You go on a nice winning streak. You eventually begin facing opponents of higher skill level. It's not the MM or Elo trying "even out" your W/L records, it's simply your Elo on a steady increase until eventually you're playing against opponents "better" than you skill wise.

it's like equating a college football player being drafted to the pros and realizing that the speed of the game has increased dramatically. It's not that the player has gotten "bad", it's that he is now playing with others at the same or higher skill level than he is at
on the entire team
as opposed to facing off against 1-2 superstars on the opposing team.

Same principle here. The MM keeps lining up opponents and nobody complains about the 10 game win streak they just came off of before going on that 12 game losing streak.

now that's not to say the MM and Elo couldn't use some work. There are still the issues of new players no longer being considered "new" after only 25 matches.
Elo being based on the team's effort as opposed to individual effort (which is why you see players with ridiculous KDR but mediocre W/L stats).
Elo is based on individual skill, it's not meant to be a "team" parameter. When you start taking averages of Elo, you've borked the whole system.
Player A = 200 Elo
Player B = 800 Elo
Average Elo = 500
That means Player B is probably going to do quite well and those players are getting the 600+ damage and multiple kills. Meanwhile, Player B is playing well above their Elo and the ones doing sub 200 damage and getting 0 kills.
Now toss in the ability to game Elo because it's set up by weight class instead of individual mech and players have the ability to "throw it off" even more.

I know a few players who keep their Locusts around just so they can have a "crappy" mech to pull out and keep their light Elo down. They don't care about that mech's stats so going in and doing poorly doesn't bother them. This drops their Elo for the entire weight class so when they pull out that optimized Jenner, they go in (at the lower Elo rating) and clean house. Then when they start losing, they switch back to the Locust.

These are the problems with how PGI has implemented Elo, not with the Elo system itself.


yes they stated that

You know, I never really thought about it until you put it that way.

Food for thought: What if everyone had a different Elo for each chassis, as opposed to weight class?

Yeah, I know that doesn't get rid of the whole "done with my first 25, put me in with the big boys" problem...nor does it address the MM's issue with thinking a Battlemaster is the same thing as a Dire Wolf but....huh.

That might actually help the problem a bit.

[Sandpit]

Willard Phule, on 03 September 2014 - 09:39 AM, said:

You know, I never really thought about it until you put it that way.

Food for thought: What if everyone had a different Elo for each chassis, as opposed to weight class?

Yeah, I know that doesn't get rid of the whole "done with my first 25, put me in with the big boys" problem...nor does it address the MM's issue with thinking a Battlemaster is the same thing as a Dire Wolf but....huh.

That might actually help the problem a bit.

I've suggested that for a long time now. Then stop averaging team Elo. Place players in a team with players that are of similar skill levels according to Elo instead of putting you (with an Elo of say 3000) in with a player who has an Elo of 1000 to average a team Elo of 2000.

The biggest problem I see with who Elo is implemented here is the whole averaging thing. That's exactly why you wind up with players well under your personal skill level and well under the personal skill level of many of your opponents.

If they gave individual mech Elo as well as stopped the practice of using the average Elo for the team, it would alleviate a LOT of that. It wouldn't make it extinct due to release valves but it would mean the difference between "15 games in a row" and "1 out of every 5 or so matches"

[MrMadguy]

Edited by MrMadguy, 03 September 2014 - 10:51 AM.

[pcunite]

What scares me is not that the match maker is bad ... but rather that PGI may have introduced Ai players. They are not real! When spectating them they play as if toddlers were using joysticks. They cannot be real people and they don't respond to questions.

Edited by pcunite, 03 September 2014 - 10:40 AM.

[BLOOD WOLF]

pcunite, on 03 September 2014 - 10:04 AM, said:

What scares me is not that the match maker is bad ... but rather that PGI is introducing Ai players. They are not real! When spectating them they play as if toddlers were using joysticks. They cannot be real people and they don't respond to questions.

They are introducing AI players?

[Screech]

Since Elo has been out never had an account under 50% win ratio, 100% solo and few thousand drops. Since the segregation it has gotten even better. Some games you have to carry, some you get carried. But blaming continual failure on the MM is fairly silly at this point in the solo queue.

[Willard Phule]

Sandpit, on 03 September 2014 - 09:53 AM, said:

I've suggested that for a long time now. Then stop averaging team Elo. Place players in a team with players that are of similar skill levels according to Elo instead of putting you (with an Elo of say 3000) in with a player who has an Elo of 1000 to average a team Elo of 2000.

The biggest problem I see with who Elo is implemented here is the whole averaging thing. That's exactly why you wind up with players well under your personal skill level and well under the personal skill level of many of your opponents.

If they gave individual mech Elo as well as stopped the practice of using the average Elo for the team, it would alleviate a LOT of that. It wouldn't make it extinct due to release valves but it would mean the difference between "15 games in a row" and "1 out of every 5 or so matches"

And I don't see why it's completely impossible to implement.

According to PGI, they decided not to handle LRMs in a "tabletop" manner (ie: groups of 5 assigned to random locations like Streaks) but to simply "roll" for each individual missile. Ok, makes sense, you let the computer do all the work faster than we could roll dice.

What would go hand in hand with that is a different balancing system than 3/3/3/3. Under the current system, an Awesome or Battlemaster is considered the same as a Dire Wolf. It needs to be split into 20 ton (or less) brackets.

[Duke Nedo]

Sandpit, on 03 September 2014 - 09:53 AM, said:

I've suggested that for a long time now. Then stop averaging team Elo. Place players in a team with players that are of similar skill levels according to Elo instead of putting you (with an Elo of say 3000) in with a player who has an Elo of 1000 to average a team Elo of 2000.

The biggest problem I see with who Elo is implemented here is the whole averaging thing. That's exactly why you wind up with players well under your personal skill level and well under the personal skill level of many of your opponents.

If they gave individual mech Elo as well as stopped the practice of using the average Elo for the team, it would alleviate a LOT of that. It wouldn't make it extinct due to release valves but it would mean the difference between "15 games in a row" and "1 out of every 5 or so matches"

...still, that does not include a mechanism that can give streaks, only uneven matches. What could give a streak I guess, is if one suddenly could jump +1000 elo and end up where there are very few players, then you could expect to be balanced by several bad players several matches in a row. Otherwise, by slowly gaining elo nothing like that should happen. Still curious what is going on....

ps. elo for each chassi would mean that you would have to play all those hundreds of games for each chassi to arrive at your elo level... I guess that's why they settled for classes. To isolate your individual contribution to the result in a 12 v 12 game takes many games....