105 replies to this topic

[MrMadguy]

Savage Wolf, on 29 February 2016 - 02:12 AM, said:

Proven false by a match result screen? How can that in any prove or disprove snowballing? Snowballing is a process, not a result.

There is big and easily distinguishable difference between snowball effect due to some noobish players dying early and coordinated stomp. In first case you lose due to some players dying early, but you feel, that remaining players have done everything, they could do. In second case you simply feel, that the difference in team skills is so enormous, that nobody on your team could do anything. For example your entire team just hides or plays too passively - you shoot at enemies and they just stand behind and do nothing, so you're 1vs12, and enemy team really looks like coordinated premade - it pushes, flanks, surrounds, ambushes. And when you get into such skilled team - you feel, that they could win this match even without you. You just stood behind deathball and fired some lurms - haven't even shot any 'Mech via direct fire. That's what my screenshot shows - two Assault 'Mechs were AFK for 90% of match duration, but enemy team sill was stomped. Snowball effect happens, cuz 1-2 enemy 'Mechs die early? Yeah? And snowball effect can't happen with enemy team, when 2 Assault are AFK on yours? Right?

p4r4g0n, on 29 February 2016 - 02:56 AM, said:

People keep saying or implying that it is "simple" to improve PSR but never actually suggest anything credible.

The fact that you can make tons of money if you create a perfect player rating and matchmaking system would appear to be a great incentive to do so but I still have not heard of any game / competition that has attained this goal.

Maybe that is a clue that it is not "simple"?

Yeah, see my sig too. The fact, that you don't like solutions, doesn't mean they aren't exist.

Sjorpha, on 29 February 2016 - 03:09 AM, said:

You are 100% right about this, the most accurate system would only measure your long term tendency to win matches.

The strange thing is that people here don't understand how measuring trends through isolation of constants work, and are stuck in a delusion that winning and losing can't be used because they are staring themselves blind looking at specific match screens thinking "I shouldn't go down/up in this match, omg!" without understanding how completely irrelevant it is.

They also don't seem to understand that using wins and losses to measure trends is completely different from win/loss ratio, and therefore believe that since win/loss ratio is a bad metric therefore wins and losses must also be bad for measuring skill trends.

Making both or either of these mistakes of course makes it impossible to understand how ELO, PSR or any other matchmaking system works, which is why we get the ******** comments like "It measures wins not skills!!!" and other such nonsense all the freaking time.

I have tried to explain this to no avail, and so far I haven't heard many answers showing any semblance of understanding these relatively simple math, except from people who already understood it before, and this is why PGI should completely ignore any and all forum discussions on this subject.

And this is wrong. Matchmaker is intended to provide as balanced matches, as it can, with/against players, who have as close skill levels to yours, as possible. W/L = 1 means nothing, because 10 stomp wins + 10 stomp loses => W/L = 1 too, while proper MM should provide 20 balanced 50/50 matches instead. It allows players to carry and to be carried. You perform poorly, but your team still wins. Tier 1 + Tier 3 vs Tier 1 + Tier 3 match seems to be balanced. But not for Tier 3 players - Tier 1 players are those, who determine the results of match and for Tier 3 players their W/L stops representing their real skill level - they start to be just punching bags, being carried towards Tier 1 by Tier 1 players.



Performance of 'Mechs, I've been playing recently:


P.S. Why do I have to explain and prove it in every single thread, that appears every day?

Edited by MrMadguy, 29 February 2016 - 03:52 AM.

[Thunderbird Anthares]

PSR is NOT supposed to be a win/loss ratio - its supposed to be a skill rating

a true skill rating system would still DROP you even on a winning match if you had a bad game - yet that happens only when you face scout 2 crack shot atlases 30 seconds into the match

right now, even if you have your barely average game and you win, you go up.... and if you have a above average (but not amazing damage wise) game and you lose, you go down and break even

that is not how a skill rating works - its supposed to measure your performance, and performance measurement in the true sense would not let you go up on a win if you had a crap match

[Savage Wolf]

MrMadguy, on 29 February 2016 - 03:18 AM, said:

There is big and easily distinguishable difference between snowball effect due to some noobish players dying early and coordinated stomp. In first case you lose due to some players dying early, but you feel, that remaining players have done everything, they could do. In second case you simply feel, that the difference in team skills is so enormous, that nobody on your team could do anything. For example your entire team just hides or plays too passively - you shoot at enemies and they just stand behind and do nothing, so you're 1vs12, and enemy team really looks like coordinated premade - it pushes, flanks, surrounds, ambushes. And when you get into such skilled team - you feel, that they could win this match even without you. That's what my screenshot shows - two Assault 'Mechs were AFK for 90% of match duration, but enemy team sill was stomped. Snowball effect happens, cuz 1-2 enemy 'Mechs die early? Yeah? And snowball effect can't happen with enemy team, when 2 Assault are AFK oo yours? Right?

All you have is anecdotal experiences you have? Nothing but your subjective assessment of how skilled your skill, your team mates and the enemy team. And in the end, all you try to conclude is that the teams are not even. That says nothing about how the game is when the teams are equal.
If the snowball effect exist and you have equal teams, then it might still result in a stomp. If those same teams had several matches, there would be many stomps, but on both sides.

I say the snowball effect exist is based on logic. Because losing one mech means one less mech to shoot at the enemy team. It's averagely only 1/12 of the team's firepower, but for each mech lost the gap increases. All the tactics of this game, the deathball, pushing at the right time, is based on the snowball effect.
There is no rubberbanding mechanic in this game, there is no way to regain strength. You can only lose strength.

[Sjorpha]

MrMadguy, on 29 February 2016 - 03:18 AM, said:

And this is wrong. Matchmaker is intended to provide as balanced matches, as it can. W/L = 1 means nothing, because 10 stomp wins + 10 stomp loses => W/L = 1 too, while proper MM should provide 20 balanced 50/50 matches instead. It allows players to carry and to be carried. You perform poorly, but your team still wins.

I make a post explaining why measuring trends through wins and losses is completely different from win/loss ratio, and how that misunderstanding prevents people from understanding matchmaking systems.

You respond by displaying the exact same confusion I just described.

The timing is almost too good to be true, but thanks for proving my point (again).

Quote

P.S. Why do I have to explain and prove it in every single thread, that appears every day?

Because bad arguments don't improve through repetition, and because your proofs are irrelevant because of the misunderstanding you displayed above.

You are using win/loss ratio and specific match screens without understanding why they are both irrelevant to measuring a players current tendencies to win or lose.

Here's a fact to chew on: The PSR/ELO/whatever MM system doesn't even know your win/loss ratio. It's not a part of the calculation in any way shape or form. It uses wins and losses and weighs that in some way, for ELO the weighing is a prediction of match result, in PSR it's match score, but the principle is the same and neither has anything to do with win/loss ratio.

Edited by Sjorpha, 29 February 2016 - 04:03 AM.

[Savage Wolf]

MrMadguy, on 29 February 2016 - 03:18 AM, said:

Yeah, see my sig too. The fact, that you don't like solutions, doesn't mean they aren't exist.

Because your link is only a slight adjustment to skill rating creep. It does not address the focus that PSR have on damage and kills in any way. It does nothing to address all the contributions to victory that cannot be measured and so is not rewarded.
PSR will still reward people who focus on damage and kills at the expense of objectives despite the fact that these are bad players and should drop in rating, yet makes them rise.

And there is a solution that I mentioned a couple of times. It's called ELO based solely on wins and losses. It's not great, but it cannot be done any more accurately. It would also enforce people's attempts to actually win instead of just getting kills.

[Thunderbird Anthares]

thats not subjective or anecdotal - that happens, and you know it, so stop pretending otherwise

the fact that stomps happen at all is a clear result of PSR not performing its intended function

[MrMadguy]

Savage Wolf, on 29 February 2016 - 03:52 AM, said:

All you have is anecdotal experiences you have? Nothing but your subjective assessment of how skilled your skill, your team mates and the enemy team. And in the end, all you try to conclude is that the teams are not even. That says nothing about how the game is when the teams are equal.
If the snowball effect exist and you have equal teams, then it might still result in a stomp. If those same teams had several matches, there would be many stomps, but on both sides.

I say the snowball effect exist is based on logic. Because losing one mech means one less mech to shoot at the enemy team. It's averagely only 1/12 of the team's firepower, but for each mech lost the gap increases. All the tactics of this game, the deathball, pushing at the right time, is based on the snowball effect.
There is no rubberbanding mechanic in this game, there is no way to regain strength. You can only lose strength.

I have been playing this game since Open Beta - I can distinguish unlucky team from unskilled one. Experienced player don't even need to spectate to do it - everything can be seen on minimap. If you see, that enemy team ambushes and you are the only one, who sees it - then you know, that your team is full of noobs without situational awareness, who can't even read map (or play in 3PV).

Sjorpha, on 29 February 2016 - 03:56 AM, said:

I make a post explaining why measuring trends through wins and losses is completely different from win/loss ratio, and how that misunderstanding prevents people from understanding matchmaking systems.

You respond by displaying the exact same confusion I just described.

The timing is almost too good to be true, but thanks for proving my point (again).



Because bad arguments don't improve through repetition, and because your proofs are irrelevant because of the misunderstanding you displayed above.

You are using win/loss ratio and specific match screens without understanding why they are both irrelevant to measuring a players current tendencies to win or lose.

Here's a fact to chew on: The PSR/ELO/whatever MM system doesn't even know your win/loss ratio. It's not a part of the calculation in any way shape or form. It uses wins and losses and weighs that in some way, for ELO the weighing is a prediction of match result, in PSR it's match score, but the principle is the same and neither has anything to do with win/loss ratio.

I see empty words only - nothing, but subjective opinion. I at least provide screenshots with profs. My performance is extremely terrible, cuz game is almost unplayable for me and despite of that my W/L = 1 and I almost advanced into Tier 2 yesterday.

Edited by MrMadguy, 29 February 2016 - 04:17 AM.

[Savage Wolf]

Thunderbird Anthares, on 29 February 2016 - 03:44 AM, said:

PSR is NOT supposed to be a win/loss ratio - its supposed to be a skill rating

a true skill rating system would still DROP you even on a winning match if you had a bad game - yet that happens only when you face scout 2 crack shot atlases 30 seconds into the match

right now, even if you have your barely average game and you win, you go up.... and if you have a above average (but not amazing damage wise) game and you lose, you go down and break even

that is not how a skill rating works - its supposed to measure your performance, and performance measurement in the true sense would not let you go up on a win if you had a crap match

Yes, it's supposed to measure an individual players contribution to victory, even if actually winning or losing. However, you cannot do this in any meaningful way. So the entire basis of the PSR is bust and a fantasy that will never come true. So trying to fix PSR is a fools errand.

[Savage Wolf]

Thunderbird Anthares, on 29 February 2016 - 03:57 AM, said:

thats not subjective or anecdotal - that happens, and you know it, so stop pretending otherwise

I know it happens. That's not the subjective part. The subjective part is the assessment of who had skill and who didn't. He only have his own experience of the match and the match results to go by. The fact that he believes he can make any judgement call at all just showes that he does not know how much bias is involved.

Thunderbird Anthares, on 29 February 2016 - 03:57 AM, said:

the fact that stomps happen at all is a clear result of PSR not performing its intended function

If the snowball effect exist as part of the game mechanics, stomps will happen regardless of how good the matchmaking could ever become. So unless you can give any arguments as to why the snowball effect is not part of the game mechanics, you cannot make the argument that the stomps are solely to blame on the matchmaking.

[Sjorpha]

Thunderbird Anthares, on 29 February 2016 - 03:57 AM, said:

the fact that stomps happen at all is a clear result of PSR not performing its intended function

No it isn't.

The intended function of PSR is to find teams of equal skill level, it isn't to prevent stomps.

Yes, that is two different things.

Uneven teams can be a reason behind a stomp, but there are several other reasons. For example a low standard deviation of skill level within the teams (as in for example a pure tier 2 game) makes a stomp more likely while a high standard deviation of skill makes it less likely. The reason for this is that a team with a few good players can afford to have their bad players killed and still carry, while a team of even players can't afford a numerical disadvantage.

The best way to reduce stomps would be to have the teams equal to each other but with very high standard deviation of skill, with tier 1-5 players on both teams. That would reduce stomps, but it would also be very frustrating because the good players would completely dominate and the bad would have little chance to perform. It would be a game of heroes and cannon fodder on both sides. Most players prefers to play with and against equals, but the downside of that is snowballing and stomps.

TL:DR: stop obsessing over stomps, they don't prove what you think they prove.

Edited by Sjorpha, 29 February 2016 - 04:15 AM.

[Lily from animove]

PSR,awesomes, Epeen size.

The name of what you use is rather negliable. Important is only the system behind it. And according to Russ townhall about PSR I doubt we get an improvement, because he said something about winning should be more part of it. Yeah GG, thats totally NOT how any system will ever rate "skill" because none of the system PGI currently made took into accound a players effort towards the win. Only if this is measured right will we ever get a correct "rating" but as long as the "dies soon derp" and the "carry hard" guy just get ups only because they both win th esystem will never properly work.

Edited by Lily from animove, 29 February 2016 - 04:17 AM.

[MrMadguy]

Stop telling us BS. This ISN'T snowball effect - this is terrible NASCAR team, stomped by good one:

Whole enemy team was there (some 'Mechs, like D-DC, were covered by ECM - I'd like to make better screenshot, but I was about to die), but half of players were facing the wrong direction.

Edited by MrMadguy, 29 February 2016 - 04:33 AM.

[Oncoshi]

El Bandito, on 28 February 2016 - 11:34 PM, said:

I don't want to group with complete newbies who doesn't even know where the "lock" button is, or shooting SRMs against target beyond 400 meters. Also the reason why I do not want trial mechs allowed in CW queue. Small groups have tonnage advantage in group queue so there is that.

Sorry but there are so many T1 and T2 players out there who doesnt know the "lock" button....

[Savage Wolf]

MrMadguy, on 29 February 2016 - 04:26 AM, said:

Stop telling us BS. This ISN'T snowball effect:

Whole enemy team was there (some 'Mechs, like D-DC were covered by ECM), but majority of players were facing the wrong direction.

The logic is there in the game mechanics. Any strategy player would be able to see it. It's a numbers game from that perspective and stomps are a natural consequence of numbers escalating as a gap increases. If this is BS then you should have no trouble giving me a counterargument instead of salt.

Your warstories are but single datapoints from a biased source with limited information. If you should prove anything this way we would need 1000 or more matches and not just the ones that you pick out but all of them. And then we would still only have a result to look at, we still wouldn't know exactly how this result occurred.

Even if matchmaking is creating stomps, it wouldn't prevent snowballing from also doing so. So examples of matchmaking creating stomps would not exclude the existence of examples with stomp where teams were equal. So you can give me all the screenshots in the world of bad matchmaking, it proves nothing about the existence of the snowball effect.

[MrMadguy]

Savage Wolf, on 29 February 2016 - 04:40 AM, said:

The logic is there in the game mechanics. Any strategy player would be able to see it. It's a numbers game from that perspective and stomps are a natural consequence of numbers escalating as a gap increases. If this is BS then you should have no trouble giving me a counterargument instead of salt.

Your warstories are but single datapoints from a biased source with limited information. If you should prove anything this way we would need 1000 or more matches and not just the ones that you pick out but all of them. And then we would still only have a result to look at, we still wouldn't know exactly how this result occurred.

Even if matchmaking is creating stomps, it wouldn't prevent snowballing from also doing so. So examples of matchmaking creating stomps would not exclude the existence of examples with stomp where teams were equal. So you can give me all the screenshots in the world of bad matchmaking, it proves nothing about the existence of the snowball effect.

Snowball effect argument would be true, if balanced matches wouldn't happen and wouldn't have happened in the past. But the simple fact, that balanced matches exist somewhere and that in the past there was a time, when all matches were balanced, suggests, that something could be done to make them happen more often. It often feels like, that it would be enough to just shuffle teams a little bit to turn stomp into more balanced match.

Comebacks are possible. I saw matches, where last 3-4 'Mechs stomped enemy team, that was winning 8:4. It also happens pretty often, that while you alive, match is even, but when you die, it seems, like you was the only skilled player, who carried this match - your team collapses almost immediately and when you spectate rest players on your team, you see, that they are complete noobs, who don't even know, what they're doing. I saw hundreds of such matches. And this leads me to a conclusion, that snowball effect doesn't exist - everything depends on difference of skill levels only. If one team has 2-4 more skilled players - they can easily stomp entire enemy team.

Edited by MrMadguy, 29 February 2016 - 05:19 AM.

[Kyynele]

MrMadguy, on 29 February 2016 - 04:54 AM, said:

in the past there was a time, when all matches were balanced

No there wasn't. Sorry.

This is the oldest screenshot I found, from 2013. But complete stomps happened in 8 vs 8, too.

[Savage Wolf]

MrMadguy, on 29 February 2016 - 04:54 AM, said:

Snowball effect argument would be true, if balanced matches wouldn't happen and wouldn't have happened in the past. But the simple fact, that balanced matches exist somewhere and that in the past there was a time, when all matches were balanced, suggests, that something could be done to make them happen more often. It often feels like, that it would be enough to just shuffle teams a little bit to turn stomp into more balanced match.

It is clear that you will not see anything but close matches as balanced, but this does not need to be true. You can easily have a balance based on snowballing. If two teams battle 1000 times and all are stomps but both teams won 500 times, is it not balanced then? Are then teams not equal then?

And if balanced matches disprove the snowball effect, then they also disprove any trouble with the matchmaker. Just because a snowball effect exist does not mean that it will snowball all the time. Many matches only snowball at the end resulting in 12 - 6 matches.

Bad matchmaking could potentially amplify the effects of the snowball effect and so good matchmaking would lessen it, but never remove it. Higher skilled players will still see stomps because it's still part of the tactics. A good push could mean the team gets the snowball effect rolling and stomps over the other team. But if it was a bad call the other team knows how to punish that mistake and then the snowball goes the other direction.

All in all, if you do not like stomps, I would not recommend this game. It's part of the core gameplay and cannot be fixed without changing the game drastically. I personally have no problem with the snowballing effect, I incorporate it into my tactics.

[El Bandito]

Oncoshi, on 29 February 2016 - 04:29 AM, said:

Sorry but there are so many T1 and T2 players out there who doesnt know the "lock" button....

Nah, unlike low tiers, those who do not lock in T2 and T1 are just being bad.

Just wish this game reverts back to 8v8 so snowballing will become not as hard...

Edited by El Bandito, 29 February 2016 - 05:44 AM.

[Savage Wolf]

El Bandito, on 29 February 2016 - 05:41 AM, said:

Just wish this game reverts back to 8v8 so snowballing will become not as hard...

Snowballing in 8v8 is even worse as each mech is a bigger percentage of the teams. If you want less snowballing, you need bigger teams.
Not that it will ever remove snowballing. The mechanics would still be there.

[El Bandito]

Savage Wolf, on 29 February 2016 - 05:52 AM, said:

Snowballing in 8v8 is even worse as each mech is a bigger percentage of the teams. If you want less snowballing, you need bigger teams.
Not that it will ever remove snowballing. The mechanics would still be there.

Nope, Snowballing was definitely less of an issue during 8v8 days, since less mechs would be focusing an exposed enemy--and that goes double for LRMs. Also, there were less deathballing and more flanking--Frozen City tunnel, for example were used much more. There were also lot more 1v1 fights happening. Finally, it allowed good pilots to make much bigger difference for their team, which is what I like.