Ranked matchmaking coming to Dota 2 - Page 17

problem with these rating systems is that they dont start nubs off at 0 they start them in the middle - well hon did and it made mmr games stupid unless you went in playing at way above average.

You are trying to argue that a small difference in skill will show through that statistical noise. I loudly cough and mutter 'bollocks'. You have no idea what you are talking about, you have no figures or anything. You are talking about elo - which is a system for 1v1 games in the context of a team of individuals with elo scores. That is not what elo was designed for. Elo *requires* 1v1 to make sense (a fixed team would be an example of a 1). You want to move into MMR its the same deal - sure people have tried to adapt it and the raft of issues are all the criticisms you have labelled bullshit. You are comparing for instance starcraft 1v1 to a team game 5v5 and are assuming all your assumptions still hold. Sorry you are wrong.

You can say people are factually incorrect as much as you like. Bold does not make you right.

This is my experience of at least 5000 games of hon from start of its beta till dota2 beta started:
You start playing at 2.5k and you go down. All is good - you have great games because everyone is same skill level you slowly work your way up. My point is that at this point everyone is same level and you are having good games This is when you hit 1000 mmr. When you start winning these games it is because you are better and can make an impact. It is good and fun.

In HON the skill level at 1500 was better than 2500 because of consistency. People had played enough to understand the basics of the game the feeders hadnt made it down that far and people were bad at the game instead of completley lost figuring out why 2 boots dont make you faster than 1 when you have 4 legs.. They were down there because they were playing lots of different heroes and not just pub stomp.

You get better and get back upto 2500 hit nubs again and you have no idea how much noise that adds in terms of wasted games and rage. When you lose a game it is about the same as 3-4 wins in terms of points made and lost - because when you have had a streak of wins certainty in your rating is low and a correction is severe. So you bounce down fast - and repeat.

Why do you bounce down? Because you need to win maybe 5-10 games to get past the minefield but there is a 50% of autolosing/autowinning due to a feeder which are in 90% of the games. Also remember this is someone who plays a lot of games ... all ranking systems carry significant inertia for these players. The occasional player on the other hand has rapid rank changes.

The fact is i am not high skill ... games with a feeder on one side (and i dont mean someone who got outclassed by a lane matchup or a player - i mean someone who is just terrible) are almost impossible for me to turn around and win because to do that you have to be REALLY fucking good.

When you get to that middle range you skill doesnt actually matter because noone in the team is of sufficiently high skill as everyone is mid skill to counter balance the gigantic weight of fail.

So no, your argument about teammates in mmr does fail. You are assuming a game that ends with a score and so small differentials can show. That is not the case. The result of the game determines point gain or loss and the actions of 4 people who are slightly better are nowhere near enough to compensate for a feeder.

Sure yes statistically the chances of having feeders on each side balances out. The problem is you need to be able to show that inside all that noise you can still make out the signal of your effect on the game. Even then you still need to show that MMR is a subtle enough score system to be able to discern those small differences over the noise. Just because statistical analysis can show someone is slightly stronger in a situation where the win and loss is not determined by them does NOT imply that the scoring system will not award so much influence to the noise that the skill part will have any discernible effect. Then once you have shown that you then need to put a figure on the number of games to iron out this variance. Your problem is that a) dota is a very landslidey game. Its easy to push an advantage into a bigger one and b) The permutations in this game are staggering. Also previous implementations (hon) made losing = about 4-5 wins which really diminishes any signal you have against the statistical noise.

You are better off finding a group of friends to team stack with to get you past that volatile section and then start playing solo again.



You would however be correct if you could score the game so that the small differentials could show. This is what valve were talking about trying to figure out, using in game statistics to help weight peoples mmr changes after games - so the pure result wasnt the only factor.

You say large enough sample sizes ... what would a large enough sample size be? Seriously work it out and tell me.
It will not be possible to get a sample size large enough. About 100 heroes is a shit ton of permutations for 5v5 yet alone getting a large enough number of games with each one. I have literally played about 10,000 games of dota and there are still matchups that catch me out.

So lets assume we can get a large enough sample size ...
you are still full of shit because you are assuming that with a large enough sample you can detect something that is dwarfed by noise ... the noise is feeders in the middle of the pack. If they feed game is over. You are talking middle of pack ... noone has the individual skill to counter a feeder at this skill level or several notches above it.

Because the non feeders impact on the game result is so low the win loss history of people represents the statistical noise of how they were matched up. Sure once you pass this band of noise mmr once again becomes meaningful - but there is this bit in the middle of about 1000mmr that is pure randomness.

You probably need an analogy:
Take poker ... take very successful cash game players.
Remove their bankrolls and start them back off with very little like when they started 5 buyins or so - even 20.
they will be the first to tell you how likley they are to go broke. If you start off on a cooler it takes that much more to get back into it because the statistical noise is so high. But guess why poker is different? You can choose how much you lose in the hand. It isn't black and white - unlike mmr and elo systems. Really most profit in poker comes first from minimising your losses and then beginning to tweak your wins.

You NEED luck to get out of these areas is my point - and then you need the skill to stay out. Dont get me wrong, if you are bad you wont get out, but that does not mean being better than the skill level of these areas imples you will get out. Your point is basically saying system is broken but if you play enough 2 events will converge 1) you will have the good fortune to get out of the area and 2) you will stay there due to your skill. The problem is that point 1 is what people are complaining about not the second.

When players do eventually get past them sure they have earnt it, however that doesn't change the fact that it was a statistical aberation that let them move on. Good for them, when i got mine in hon it was a godsend, but it took months.

In SC2 people start at 0 and arnt in teams (team matchmaking is hilarious - at least you can take resources of deadweight though). MMR in hon at least didn't start from 0 which was the HUGE problem. The relevance of this is you are making a lot of assumptions that are not necessarily correct and are invoking the god of statistics in an apallingly naive way.

problem with these rating systems is that they dont start nubs off at 0 they start them in the middle - well hon did and it made mmr games stupid unless you went in playing at way above average.

You are trying to argue that a small difference in skill will show through that statistical noise. I loudly cough and mutter 'bollocks'. You have no idea what you are talking about, you have no figures or anything. You are talking about elo - which is a system for 1v1 games in the context of a team of individuals with elo scores. That is not what elo was designed for. Elo *requires* 1v1 to make sense (a fixed team would be an example of a 1). You want to move into MMR its the same deal - sure people have tried to adapt it and the raft of issues are all the criticisms you have labelled bullshit. You are comparing for instance starcraft 1v1 to a team game 5v5 and are assuming all your assumptions still hold. Sorry you are wrong.

You can say people are factually incorrect as much as you like. Bold does not make you right.

This is my experience of at least 5000 games of hon from start of its beta till dota2 beta started:
You start playing at 2.5k and you go down. All is good - you have great games because everyone is same skill level you slowly work your way up. My point is that at this point everyone is same level and you are having good games This is when you hit 1000 mmr. When you start winning these games it is because you are better and can make an impact. It is good and fun.

In HON the skill level at 1500 was better than 2500 because of consistency. People had played enough to understand the basics of the game the feeders hadnt made it down that far and people were bad at the game instead of completley lost figuring out why 2 boots dont make you faster than 1 when you have 4 legs.. They were down there because they were playing lots of different heroes and not just pub stomp.

You get better and get back upto 2500 hit nubs again and you have no idea how much noise that adds in terms of wasted games and rage. When you lose a game it is about the same as 3-4 wins in terms of points made and lost - because when you have had a streak of wins certainty in your rating is low and a correction is severe. So you bounce down fast - and repeat.

Why do you bounce down? Because you need to win maybe 5-10 games to get past the minefield but there is a 50% of autolosing/autowinning due to a feeder which are in 90% of the games. Also remember this is someone who plays a lot of games ... all ranking systems carry significant inertia for these players. The occasional player on the other hand has rapid rank changes.

The fact is i am not high skill ... games with a feeder on one side (and i dont mean someone who got outclassed by a lane matchup or a player - i mean someone who is just terrible) are almost impossible for me to turn around and win because to do that you have to be REALLY fucking good.

When you get to that middle range you skill doesnt actually matter because noone in the team is of sufficiently high skill as everyone is mid skill to counter balance the gigantic weight of fail.

So no, your argument about teammates in mmr does fail. You are assuming a game that ends with a score and so small differentials can show. That is not the case. The result of the game determines point gain or loss and the actions of 4 people who are slightly better are nowhere near enough to compensate for a feeder.

Sure yes statistically the chances of having feeders on each side balances out. The problem is you need to be able to show that inside all that noise you can still make out the signal of your effect on the game. Even then you still need to show that MMR is a subtle enough score system to be able to discern those small differences over the noise. Just because statistical analysis can show someone is slightly stronger in a situation where the win and loss is not determined by them does NOT imply that the scoring system will not award so much influence to the noise that the skill part will have any discernible effect. Then once you have shown that you then need to put a figure on the number of games to iron out this variance. Your problem is that a) dota is a very landslidey game. Its easy to push an advantage into a bigger one and b) The permutations in this game are staggering. Also previous implementations (hon) made losing = about 4-5 wins which really diminishes any signal you have against the statistical noise.

You are better off finding a group of friends to team stack with to get you past that volatile section and then start playing solo again.



You would however be correct if you could score the game so that the small differentials could show. This is what valve were talking about trying to figure out, using in game statistics to help weight peoples mmr changes after games - so the pure result wasnt the only factor.

You say large enough sample sizes ... what would a large enough sample size be? Seriously work it out and tell me.
It will not be possible to get a sample size large enough. About 100 heroes is a shit ton of permutations for 5v5 yet alone getting a large enough number of games with each one. I have literally played about 10,000 games of dota and there are still matchups that catch me out.

So lets assume we can get a large enough sample size ...
you are still full of shit because you are assuming that with a large enough sample you can detect something that is dwarfed by noise ... the noise is feeders in the middle of the pack. If they feed game is over. You are talking middle of pack ... noone has the individual skill to counter a feeder at this skill level or several notches above it.

Because the non feeders impact on the game result is so low the win loss history of people represents the statistical noise of how they were matched up. Sure once you pass this band of noise mmr once again becomes meaningful - but there is this bit in the middle of about 1000mmr that is pure randomness.

You probably need an analogy:
Take poker ... take very successful cash game players.
Remove their bankrolls and start them back off with very little like when they started 5 buyins or so - even 20.
they will be the first to tell you how likley they are to go broke. If you start off on a cooler it takes that much more to get back into it because the statistical noise is so high. But guess why poker is different? You can choose how much you lose in the hand. It isn't black and white - unlike mmr and elo systems. Really most profit in poker comes first from minimising your losses and then beginning to tweak your wins.

You NEED luck to get out of these areas is my point - and then you need the skill to stay out. Dont get me wrong, if you are bad you wont get out, but that does not mean being better than the skill level of these areas imples you will get out. Your point is basically saying system is broken but if you play enough 2 events will converge 1) you will have the good fortune to get out of the area and 2) you will stay there due to your skill. The problem is that point 1 is what people are complaining about not the second.

When players do eventually get past them sure they have earnt it, however that doesn't change the fact that it was a statistical aberation that let them move on. Good for them, when i got mine in hon it was a godsend, but it took months.

In SC2 people start at 0 and arnt in teams (team matchmaking is hilarious - at least you can take resources of deadweight though). MMR in hon at least didn't start from 0 which was the HUGE problem. The relevance of this is you are making a lot of assumptions that are not necessarily correct and are invoking the god of statistics in an apallingly naive way.