277 replies to this topic

[Hauser]

PhoenixFire55, on 23 April 2013 - 02:04 AM, said:

Your example is good and true, but irrelevant to what we discuss. I will fully agree with you that 1 person can be a deciding factor in 1 vs 1 games, 2 vs 2 games, probably even in 3 vs 3 games, but it is not so in 8 vs 8 and surely not so in 12 vs 12. This 'margin', 'difference' he makes is becoming less and less the more people you have in your teams. You can calculate this yourself if you extend your example into 8 vs 8 and 12 vs 12. Chances of a good player winning will be like 52-54%, chances of a bad player winning 46-48%. This is exactly what we see on graphs, 90% of people are there in between 46-54% wins, only people outside these boundaries are people doing the two things I've mentioned in my reply to Wispsy just above...

Because I can.

Tug of war. 12 vs 12. One million random teams. Approximate bell curve for player strengths. Average skill is 2.64

Strength:Win%
0:0.3702995
1:0.4232725
2:0.4731635
3:0.5222365
4:0.5720765
5:0.6237935
6:0.6835585

As you can see average players (2-3) win around 50%. Poor players (0-1) around 40% and good payers (4-5) around 60% and the elite (6) are closing on on 70%.

The elite and good players carry a good number of of their teams. Like wise the poor drag their team down very often. They are both the unbalancing factor in their team. As such wins and losses are just fine to judge people by, even though they play with 8 other people.

I'm not sure what graph you're talking about the only one I've seen shows elo. Not wins or losses.

PhoenixFire55, on 23 April 2013 - 02:04 AM, said:

I've just had two 8:0 roflmao stomps in a row, just like I used to have in closed beta. I can't call it quality matchmaking.

It's European afternoon right now. American morning. I don't think there are enough people playing to make any match maker work.

Edited by Hauser, 23 April 2013 - 03:25 AM.

[PhoenixFire55]

Hauser, on 23 April 2013 - 03:06 AM, said:

Because I can.

Tug of war. 12 vs 12. One million random teams. Approximate bell curve for player strengths. Average skill is 2.64

Strength:Win%
0:0.3702995
1:0.4232725
2:0.4731635
3:0.5222365
4:0.5720765
5:0.6237935
6:0.6835585

As you can see average players (2-3) win around 50%. Poor players (0-1) around 40% and good payers (4-5) around 60% and the elite (6) are closing on on 70%.

The elite and good players carry a good number of of their teams. Like wise the poor drag their team down very often. They are both the unbalancing factor in their team. As such wins and losses are just fine to judge people by, even though they play with 8 other people.

Your model also shows what I mean ... total epic fail 0-skill guy who is just afk still wins 37% of his games. And vast majority of people around your 2.64 average skill in between skill 2 and 4 win 47-57% of their games (my calculation was 46-54% no real difference).You can easily get another numbers changing your parameters, for one you did increase the number of player skill levels to 7, you can't really tell if its 7 or 3 or 78.

Another thing is that in this model (lets take your previous one with 3 skill levels):
good+good = 2+2 = 3+1 = elite+bad
While in reality there is such thing as teamwork, so in fact:
good+good = (2+2)*1.5 > (3+1)*0.8 = elite bad
What I mean is that 'goods' will agree to focus fire and stick together and such and 'bad' will just shoot his 'elite' in the back, leave him alone or better yet just go OB.

An 'elite' player will not 'carry' his team. Because if he is in a really bad team he alone can't do anything. Same time if he is in a good team everyone does well and he isn't carrying anyone.

The difference between 'elite' and 'epic fail' just isn't significant enough (37% to 68%, 47% to 57% for majority of people) which means that 'elites' will get 'epic fails' in their games quite often. Same ... I can't call this quality matchmaking, its not any different from totally random matchmaking.

[Hauser]

PhoenixFire55, on 23 April 2013 - 01:53 AM, said:

3 and 4 are false. It doesn't matter just how good/bad he performs. There is a high chance that the other team has the same new player. They both have no affect on the game outcome whatsoever. Chances of a completely new player being placed on a good team or a bad team are same. Thus chances of a completely new player to win/lose are the same. He'll be winning 50% of his game right from the start.

The chance for one new player (with average elo) to face another new player (with average elo) would depend on the population.

• Assuming there are only a few new players (with average elo) the chance of there being another new player (with average elo) on the other team is low, not high. As such a new player (with average elo) will have a greater chance to lose and go down in rank. The model shows this effectively happens.
  
• Assuming there are many new player (with average elo) the chance of there being another new player (with average elo) on the other team is high. As such the chances of winning are indeed closer to 50% however. In that scenario the averagely skilled player (with average elo) will win more often from these new players (with average elo) and go up.

So either new players migrate down or the average players migrate up. Though it would be more accurate to say that with so many new players, what used to be an averagely skilled player is in fact experienced.

Edited by Hauser, 23 April 2013 - 03:58 AM.

[Hauser]

PhoenixFire55, on 23 April 2013 - 03:48 AM, said:

You can easily get another numbers changing your parameters, for one you did increase the number of player skill levels to 7, you can't really tell if its 7 or 3 or 78.

Correct. These are presumed to be hidden in reality. I do however need some way to simulate this tug of war game to demonstrate that people who are better will out preform others when using a random team composition.

PhoenixFire55, on 23 April 2013 - 03:48 AM, said:

Your model also shows what I mean ... total epic fail 0-skill guy who is just afk still wins 37% of his games. And vast majority of people around your 2.64 average skill in between skill 2 and 4 win 47-57% of their games (my calculation was 46-54% no real difference).

I don't get what you're trying to say here.

When matching up people randomly average people can be expected to win 50% of their games by virtue of being average. They'll encounter good players and better players on their team as often as they do encounter them on the other team. They'll encounter them in combination that help them as often as it goes against them.

Just as the 0-skill guy gets lucky from time to time with his team composition. However not lucky enough to win all the time and this will still cause him to lose elo over time.

Now this is with a random match maker. What would happen if we used an elo match maker? Say this 0-skill guys is new and has an average elo. The average elo player population has a skill of 2 and there are enough players available to make a team with only average players. There are no other new players.

12vs12 1 million matches.

Strenght:Win%
0:0.0
2:0.522239

The effect is overwhelming now. The advantage the random match maker gave this guy by putting him on a good team by pure chance is gone now. He has not won a single game and as a result his elo will sink like a brick (at least until the match maker puts him in with other people).

Now this is not a realistic situation as the charts show nearly 50% of the population has played less then 50 games. These are al at average elo levels (and I agree, they shouldn't be there).

12vs12 1 million matches. 50% 0, 50% 2.

0:0.4207155
2:0.5790735

The effect is rather limited however still enough to cause people to migrate to a different elo level.

.

PhoenixFire55, on 23 April 2013 - 03:48 AM, said:

Another thing is that in this model (lets take your previous one with 3 skill levels):
good+good = 2+2 = 3+1 = elite+bad
While in reality there is such thing as teamwork, so in fact:
good+good = (2+2)*1.5 > (3+1)*0.8 = elite bad
What I mean is that 'goods' will agree to focus fire and stick together and such and 'bad' will just shoot his 'elite' in the back, leave him alone or better yet just go OB.

Agreed. The model just assumes people add their strength and does not account for synergies. I'm doing this on purpose because I want to keep the model simple. I'm not trying to simulate a mwo match, I'm trying to show that it is possible using only wins and losses to rank people using elo, even though they are playing in a team and their (elite) contribution might not always result in win.

PhoenixFire55, on 23 April 2013 - 03:48 AM, said:

An 'elite' player will not 'carry' his team. Because if he is in a really bad team he alone can't do anything. Same time if he is in a good team everyone does well and he isn't carrying anyone.

I'd like to reverse the order of what you are saying. If an elite player is in a really bad team he alone will not 'carry' his team. This is correct. Likewise if an elite player is in a really good team he doesn't have to do much. However the simulation shows that with random match making an elite (6) player will be in a winning team more often then not. He can be expected to win 68% of his games. This being the elite player and the only constant in all these winning teams we can assume that he is doing the work and carrying his team.

PhoenixFire55, on 23 April 2013 - 03:48 AM, said:

The difference between 'elite' and 'epic fail' just isn't significant enough (37% to 68%, 47% to 57% for majority of people) which means that 'elites' will get 'epic fails' in their games quite often. Same ... I can't call this quality matchmaking, its not any different from totally random matchmaking.

These win percentages are produced by a random matchmaker! You think I have time to write an elo match maker over lunch? Give me till dinner.

I keep brining this random match maker up because 1) it shows that wins can be used to measure individual contribution in a team and 2) the initial state of an elo system is equivalent to a random match maker. If a elite player has a higher chance to win he will inevitably over the course of a few matches acquire elo. Just as a new player will lose them over the course of a few games. Once the divide is big enough the match maker will avoid trying to put them in the same game. They might go up or down further until the point where they've met their match and can be served with a 50% to win.

Edited by Hauser, 23 April 2013 - 05:19 AM.

[TexAce]

So if the light blue curve is players with less than 50 matches played, it means these are fairly new players. Since the peak of the light blue curve is on a higher ELO than the dark blue one shouldn't we adjust it that way that the starting value is lower than the peak of the dark blue curve?

right now it looks as if new players begin on a too high median ELO and slowly move downwards to the correct value. I think it would be better the other way around. start at a lower ELO and move up.

[Wispsy]

PhoenixFire55, on 23 April 2013 - 01:47 AM, said:

Also, I don't know why you of all people worry bout being placed bottom of the team. Each time I've been in game with you you've been top or near top of your team in a jenner, dragon, any mech really. As I said, even with the current very imperfect game score good players end up with good game scores almost all the time, regardless of what mechs they use.

I still think that 1 man can't carry his team, I've never seen that happen. If smb claims to have 8 kills in a game then it only means his teammates contributed greatly by stripping off enemy armor, taking fire etc. One person can't swing a balance in a non-close matchup, much less so he'll be able to do it in 12 vs 12. And if its a close matchup you can't say that anyone is carrying the team, because in close matchups everyone contributes. It might happen so that in the end somebody will decide the game, but it doesn't mean he carried his team.

I disagree I have seen people other then myself carry games pretty much solo on multiple occasions (I have seen myself do it often as well but I am biased so it might not have been as much me as I thought). Yes in games when nobody hits their targets the guy actually hitting will have more damage and be higher and this is right. However when people aim for specific components and often hit where they aim the guy with the most damage is either the one with a lucky ammo explosion or the person who missed his targets the most and just spread damage randomly all over the enemy mechs. If you get a headshot just after somebody blows off a useless side torso then you will be lower on the scoreboard. The games where I do not care and just fire randomly at everything I can yeah I get high numbers and yeah it looks good. The games that I perform really well in I will almost never appear high on the scoreboard, with the right amount of positioning and accuracy you can destroy the other team whilst taking and dealing very very little damage. Then your team mops up and unless they also do it with the same accuracy then they will possibly have higher damage numbers by quite a bit.

Here is one example that actually happens of how somebody can carry a game on their own whilst accumulating almost no score at all and even if you included all of those points into the score you suggested still get bottom. First get a light mech, say a Jenner or a non ECM spider (so the enemy can target you easily). Start with some random taunt to make sure people notice you and want to kill you (i.e. your lives i claim them). Get in view of the enemy as quickly as possible and dance around until they start to chase you. String them out taking the long route back to your team so they are not all clustered together. When you have got them in some good positions lead one or two close to your team. Once you are in a range that a number of your teammates can shoot them easily (keeping in mind they may not be great shots so the closer the better) lead them out round a corner keeping them facing away from your team and just jump around keeping attention. Your team will most likely open fire on the first enemies exposed back and they will die rather quickly (if you suspect bad shots/builds make sure at least 3/4 have an easy obvious target). The enemy is unlikely to turn on them as they have spent all this time chasing you, do not know how many enemy can see them and they want that easy shot you appear to be offering finally and well 3/4 mechs of any build opening up on a back is going to cripple that mech in no time. Now go and charge at the next one who should be very close and turn him around and dance so that your team has yet another alpha to the back of a close range target and they die too. The enemy will continue pushing or your team will start to push so just keep on getting to the obvious targets first as they are nicely strung out and turning them round as your team opens up on them. If the enemy team is bad or mad enough then you can literally go through the entire game keeping your whole team at 100% regardless of their skill. You can even chose which person/small number of people to get the highest scores. Whilst this is good fun and by doing it you win almost every game your score will always be the lowest and if Elo is based on score not wins you will keep on dropping.
On the other side of that you can take 6 medium pulses on your Jenner. Then run into solo/small groups of targets and get multiple headshots in 30 seconds. You team will then push. If they are not very good then they will get loads of damage numbers finishing off the fleeing enemy mechs as they are unable to cleanly shoot out the ct. No matter what metric you base the score on unless it is kills (which people are also apparently against as it promotes killstealing?) you are finishing at the bottom of the scoreboard and even if you win every game your Elo would be hurt for it. In both these situations (both have happened on multiple occasions they are not hypothetical) clearly the other team is being outplayed and the result is a victory yet with a very low score. This is really why I believe Elo should mostly be based on wins, if you play better, you win. Basing it on score is not as good, because people can make the right calls at the right moments to pull out a victory whilst having very little score no matter the measurements used. If you cannot carry a game on your own then you are near the right Elo, I cannot carry that many games completely on my own any more but I could if I was in a game with all new players. I could probably literally 1v8 all new players tbh. The fact that I generally play against people that I cannot do that to means Elo is working I cannot see anything else to base it on other then wins though, it is all too situational. Even though those situations may not be common for every person in every game they can still easily happen and the extremely good would end up only facing the very bad. Anyway I am not sure if I have gone completely off topic or not as I have just been adding random thoughts throughout the morning but I do hope something here makes sense.

Edited by Wispsy, 23 April 2013 - 06:44 AM.

[Hauser]

TexAss, on 23 April 2013 - 05:38 AM, said:

So if the light blue curve is players with less than 50 matches played, it means these are fairly new players. Since the peak of the light blue curve is on a higher ELO than the dark blue one shouldn't we adjust it that way that the starting value is lower than the peak of the dark blue curve?

right now it looks as if new players begin on a too high median ELO and slowly move downwards to the correct value. I think it would be better the other way around. start at a lower ELO and move up.

Yes either that or putting them on an accelerated change.

[PhoenixFire55]

Wispsy, on 23 April 2013 - 06:41 AM, said:

Here is one example that actually happens of how somebody can carry a game on their own whilst accumulating almost no score at all and even if you included all of those points into the score you suggested still get bottom...

Using other peoples stupidity should not be rewarded
On a serious note yeah, I see how this can happen, but I still think that game score is a better base for Elo matchmaker then the wins/losses. Question is different, just how we can make a game score that properly rewards players actions on a battlefield apart from damage, kills, etc.

[PhoenixFire55]

Hauser, on 23 April 2013 - 04:56 AM, said:

These win percentages are produced by a random matchmaker! You think I have time to write an elo match maker over lunch? Give me till dinner.

Its rather intersting how the 'matchmaker feedback' topic turned into almost scientific discussion. I hope that PGI guys will actually find smth useful in our posts. I'd leave it to them to work on simulating current Elo and other possible variants to determine which one produces better matches tho. Unless we have the data they have we are mostly just speculating.

[Wispsy]

PhoenixFire55, on 23 April 2013 - 08:22 AM, said:

Using other peoples stupidity should not be rewarded
On a serious note yeah, I see how this can happen, but I still think that game score is a better base for Elo matchmaker then the wins/losses. Question is different, just how we can make a game score that properly rewards players actions on a battlefield apart from damage, kills, etc.

You cannot make a scoring system that will work better then replacing win/loss. It is not wrong to use win/loss, you are the constant in all your games you will eventually end up with/against similar skill, depending on how far the mm searches.

[Jonathan Paine]

I think Elo is currently changing too rapidly. If the matchmaker successfully puts two equally ranked teams into most matches, this is what happens for two 4 man teams who are equally skilled:

4x1300=5200 Elo
Win = +25 Elo x4
Loss = -25 Elo x4

New team Elo
Win: 5300
Loss: 5100

And that is one match. What if one team has a "lucky" three win streak, while the other has an "unlucky" three loss streak?
Lucky team Elo: 5500
Unlucky team Elo: 4900

At this point the lucky team will face much better teams, while the unlucky team faces much worse teams. The results are likely to be a quick 0 - 8 match for the previously lucky team, and a quick 8-0 for the previously unlucky team. This should, in the long run, lead to pretty much the same win/loss ratio for both teams as they win or lose according to their skills.

Sounds great? No. The fluctuation in Elo ratings increases the likelihood of bad teams facing good teams, which leads to more lopsided wins and losses. Personally I loathe lopsided losses, and I tend not to enjoy lopsided wins.

Solution? Halve the amount of Elo gain/loss per match. That's right:

ELO IS OP! NERF ELO!

[PhoenixFire55]

Wispsy, on 23 April 2013 - 08:29 AM, said:

You cannot make a scoring system that will work better then replacing win/loss. It is not wrong to use win/loss, you are the constant in all your games you will eventually end up with/against similar skill, depending on how far the mm searches.

Right now its about team composition more then anything. Put 8 guys of same skill into 'cheese' builds and 8 guys of same skill into stock mechs outcome is always gonna be same. At the current state of the game winning isn't about your skill (or your team skill) but rather about what mech builds you use. Not matching mech types IMO is stupid. This is my main issue with the matchmaker right now. Apart from that its whichever team has more D/C's and afk'ers or more ECM.

[hammerreborn]

Wispsy, on 23 April 2013 - 06:41 AM, said:

I disagree I have seen people other then myself carry games pretty much solo on multiple occasions (I have seen myself do it often as well but I am biased so it might not have been as much me as I thought). Yes in games when nobody hits their targets the guy actually hitting will have more damage and be higher and this is right. However when people aim for specific components and often hit where they aim the guy with the most damage is either the one with a lucky ammo explosion or the person who missed his targets the most and just spread damage randomly all over the enemy mechs. If you get a headshot just after somebody blows off a useless side torso then you will be lower on the scoreboard. The games where I do not care and just fire randomly at everything I can yeah I get high numbers and yeah it looks good. The games that I perform really well in I will almost never appear high on the scoreboard, with the right amount of positioning and accuracy you can destroy the other team whilst taking and dealing very very little damage. Then your team mops up and unless they also do it with the same accuracy then they will possibly have higher damage numbers by quite a bit.

Here is one example that actually happens of how somebody can carry a game on their own whilst accumulating almost no score at all and even if you included all of those points into the score you suggested still get bottom. First get a light mech, say a Jenner or a non ECM spider (so the enemy can target you easily). Start with some random taunt to make sure people notice you and want to kill you (i.e. your lives i claim them). Get in view of the enemy as quickly as possible and dance around until they start to chase you. String them out taking the long route back to your team so they are not all clustered together. When you have got them in some good positions lead one or two close to your team. Once you are in a range that a number of your teammates can shoot them easily (keeping in mind they may not be great shots so the closer the better) lead them out round a corner keeping them facing away from your team and just jump around keeping attention. Your team will most likely open fire on the first enemies exposed back and they will die rather quickly (if you suspect bad shots/builds make sure at least 3/4 have an easy obvious target). The enemy is unlikely to turn on them as they have spent all this time chasing you, do not know how many enemy can see them and they want that easy shot you appear to be offering finally and well 3/4 mechs of any build opening up on a back is going to cripple that mech in no time. Now go and charge at the next one who should be very close and turn him around and dance so that your team has yet another alpha to the back of a close range target and they die too. The enemy will continue pushing or your team will start to push so just keep on getting to the obvious targets first as they are nicely strung out and turning them round as your team opens up on them. If the enemy team is bad or mad enough then you can literally go through the entire game keeping your whole team at 100% regardless of their skill. You can even chose which person/small number of people to get the highest scores. Whilst this is good fun and by doing it you win almost every game your score will always be the lowest and if Elo is based on score not wins you will keep on dropping.
On the other side of that you can take 6 medium pulses on your Jenner. Then run into solo/small groups of targets and get multiple headshots in 30 seconds. You team will then push. If they are not very good then they will get loads of damage numbers finishing off the fleeing enemy mechs as they are unable to cleanly shoot out the ct. No matter what metric you base the score on unless it is kills (which people are also apparently against as it promotes killstealing?) you are finishing at the bottom of the scoreboard and even if you win every game your Elo would be hurt for it. In both these situations (both have happened on multiple occasions they are not hypothetical) clearly the other team is being outplayed and the result is a victory yet with a very low score. This is really why I believe Elo should mostly be based on wins, if you play better, you win. Basing it on score is not as good, because people can make the right calls at the right moments to pull out a victory whilst having very little score no matter the measurements used. If you cannot carry a game on your own then you are near the right Elo, I cannot carry that many games completely on my own any more but I could if I was in a game with all new players. I could probably literally 1v8 all new players tbh. The fact that I generally play against people that I cannot do that to means Elo is working I cannot see anything else to base it on other then wins though, it is all too situational. Even though those situations may not be common for every person in every game they can still easily happen and the extremely good would end up only facing the very bad. Anyway I am not sure if I have gone completely off topic or not as I have just been adding random thoughts throughout the morning but I do hope something here makes sense.

Wall of text crits thread for OMG SQUIRREL KILL IT NOW

You were mean to me Wispy. I was just trying to have a BBQ with chavette and you had to come and ruin my fun

[Wispsy]

PhoenixFire55, on 23 April 2013 - 08:53 AM, said:

Right now its about team composition more then anything. Put 8 guys of same skill into 'cheese' builds and 8 guys of same skill into stock mechs outcome is always gonna be same. At the current state of the game winning isn't about your skill (or your team skill) but rather about what mech builds you use. Not matching mech types IMO is stupid. This is my main issue with the matchmaker right now. Apart from that its whichever team has more D/C's and afk'ers or more ECM.

So somebody who builds an effective mech will win over somebody who does not and this is a bad thing? D/C can be annoying but random is random and it evens out in the end.

[Aidan]

With the current ELO system, what is the probability of an 8 pilot group composed of all light class mechs finding a match?

[hammerreborn]

Aidan, on 23 April 2013 - 10:36 AM, said:

With the current ELO system, what is the probability of an 8 pilot group composed of all light class mechs finding a match?

Fairly low, but the same goes for any 8 man group aside from certain peak hours. After about midnight EST last night we stopped getting games entirely.

[Hauser]

Jonathan Paine, on 23 April 2013 - 08:52 AM, said:

Solution? Halve the amount of Elo gain/loss per match. That's right:

ELO IS OP! NERF ELO!

lol

But I don't think it will be that bad. The average is used, not the total team value.

[PhoenixFire55]

Wispsy, on 23 April 2013 - 09:17 AM, said:

So somebody who builds an effective mech will win over somebody who does not and this is a bad thing? D/C can be annoying but random is random and it evens out in the end.

You know what 'cheese' is. Most people use it to be competitive, not to exploit it, but it remains same. A 'cheezy' team will almost always beat a non-cheese one, provided skill levels are near same.

[Jess Hazen]

remove elo and we're good

[Jonathan Paine]

I am getting the sinking feeling that the matchmaker is now more broken than... ever... Something about recognizing pretty much every enemy pilot from the top15 charts, while we are helping a couple of newer players learn to play MWO. Out Eloed, and outweighed. Took <10 seconds to find the match.

Next few matches were either us stomping people who were in trial mechs, or getting stomped by top loaded teams. What is going on?