First of all, this is not a conspiracy thread. This is not to secretly complain about how the matchmaking system is rigged or whatever else, so, please, do not talk about it here.
I've been reading a lot on the forums about matchmaking, and I couldn't find any threads that really talk about specifically improving the matchmaking system. I've seen a few posts in other threads where people mention that there seems to be a secondary system in addition to your ranking that helps determine your next match-up, but I don't know if this is true or even provable outside of taking Blizzard's word about how the matchmaking works, and this is not the main point of the thread.
So, is there a better way to decide matches in Hearthstone that any of you have thought or heard of? This would be in addition to the Ranking System used below Legend, and in addition to the MMR of Legend and Casual. If you are going to use some form of statistic to help your argument, please make it easy to be understood for other people who may see it, and most importantly, it needs to be definitive, with some form of proof.
This is not to try to say we are better than Blizzard or anything of the sort, I am just curious about other people's ideas on how it could be improved, and once again, this is not a salt or conspiracy thread, so do not post about anything of the like.
Rollback Post to RevisionRollBack
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
No I agree with you man..I feel honestly that everytime your about to hit winstreak that they match you with a counter deck type..I know there’s probably a better way to put it but I feel like it’s waaaaay to consistent for it to just be a coincidence..it’s happened with every class for me and it doesn’t matter what you do
For statistical porpoises, I'd like the matchmaking system to avoid rematching against the same opponent twice in a row if it possibly can without a long wait time because that knackers the assumptions of statistical tests that the opponent is drawn at random from the pool of all available suitable players. The reason rematches are common (especially at high ranks/late night early morning games) is because the pool of players near your rank is low and you both probably hit the Play button at about the same time after your previous match.
Rematches should be excluded as outliers from sample data since you are very likely to face the exact same deck -> the events are not independent of one another -> skews the data about distribution of classes played on ladder (for small sample sizes anyway. If you collect data from a lot of players, it becomes irrelevant).
For statistical porpoises, I'd like the matchmaking system to avoid rematching against the same opponent twice in a row if it possibly can without a long wait time because that knackers the assumptions of statistical tests that the opponent is drawn at random from the pool of all available suitable players. The reason rematches are common (especially at high ranks/late night early morning games) is because the pool of players near your rank is low and you both probably hit the Play button at about the same time after your previous match.
Rematches should be excluded as outliers from sample data since you are very likely to face the exact same deck -> the events are not independent of one another -> skews the data about distribution of classes played on ladder (for small sample sizes anyway. If you collect data from a lot of players, it becomes irrelevant).
That is all :P
I meant it more in a way of matching players against each other using more than just the fact they happen to be close in Ranking on the ladder, but I do find the mathematics of Hearthstone top be very important, even though I am unable to participate on that level because I have yet to take courses in college that could help me be able to do as such.
Rollback Post to RevisionRollBack
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
Well the only thing I can think of off the top of my head would be an algorithm like the PageRank algorithm Google used as their first search algorithm to match decks of similar quality against each other.
You'd have to give a score for each card and combinations of cards used in the deck to come up with a rank for each players deck compared to current win rates then match decks of similar rating against each other. Vicious Syndicate has charts which show how common each card is in a particular deck archetype, this could be used as basis for a score, and the win rate for the archetype would be an additional factor on top.
Well the only thing I can think of off the top of my head would be an algorithm like the PageRank algorithm Google used as their first search algorithm to match decks of similar quality against each other.
You'd have to give a score for each card and combinations of cards used in the deck to come up with a rank for each players deck compared to current win rates then match decks of similar rating against each other. Vicious Syndicate has charts which show how common each card is in a particular deck archetype, this could be used as basis for a score, and the win rate for the archetype would be an additional factor on top.
EDIT: You would use something like this to rate how similar a deck is to an existing archetype.
That leaves the quesition of what would you be aiming for. True, you can determine if deck X is a counter to archetype Y but then what do you do with it? Do you specifically force counter decks to feed off of the top decks? Kind of feels wrong to purposely punish players that way. "Oh I'm sorry, that's the wrong deck, now face this swarm of counters to ruin your rank HAHAHA!"
Note that when issues of win rates and MMR come in it's always player based, not character based. In starcraft, I would be matched up to you based on how well you win no matter what race you played. They wouldn't check to see if you were a protoss player, determine if terran had a better win rate, then decide whether me, as terran, should match up to you. in street fighter, if Ryu win more than Chun'li, they don't dump all of the Ryu's together and you don't get to drop down to the newbie zone if you chose one of the losing characters.
Matchmaking based on deck type in a genre of game designed around rock paper scissors styled balancing means effectively picking the winners and losers. It also becomes insanely easy to game (if I know that the system throws similar win rate decks together then I can just pick the deck with a win rate that encourages mirrors then tech to win the mirror).
Pesonally, I think if you are going to let the system mettle, it should be gently, just enough to soften the more extreme trends of randomness. Have it so that the game slightly favors a class different from the one you played last game. Just slightly. Thus if you've faced 3 druids in a row, and someone at your rank queued priest, you'd be pushed into the priest. If no one has, you get another druid as usual.
If you want something more, add another lean towards similar Playtime. A person who's been playing 1 hour a day for 2 weeks shouldn't be facing my 3 year old (#)$# if there's other younger folks available. No matter the deck or cards, a long-lasting player PLAYS differently from a newbie.
Beyond that I think the problems of ranked are more involved with other parts rather than matchmaking, such as the star system and how it downgrades you so heavily a tthe start of the season and how there's not much to aim for if you aren't specifically aming for legend.
PageRank style systems aren't about matching up counter decks, they are about matching decks of similar quality. Each deck would get a rating based on similarity to standard deck X which has quality Y and matching up to similarly rated decks. EDIT: You basically give each deck a rating and your deck is measured against similar decks by similarity to the reference deck. Search algorithms for the interwebz use this kind of thing to sort the results.
This could bring in a Millhouse Manastorm meta of course to throw the quality calculation off ;)
EDIT: Another problem with it would be that if there is a clearly superior deck then anyone playing it would often end up playing a mirror match. That could be amusing for a while I suppose.
This is a very interesting idea of how to improve matchmaking, but is it feasible at all? How hard/easy would it be to game the system to your needs to climb, and how difficult would it be to implement this system?
I'm still also interested in hearing other ideas of how to improve, so feel free to share any ideas you've had!
It would be feasible (just compare cards that are missing from the reference deck and adjust the score to compensate).
It can be gamed (put in deliberately bad cards and hope you don't draw them/never play them).
It's not difficult to implement if you know everyone's deck (like they do, it gets sent to the servers).
It's fast as well, which is why Google killed Yahoo as the best search engine very quickly. It basically involves doing a dot product of your decks score (which is a vector i.e. a bunch of numbers whose order is important, so multi-dimensional rather than a single number) against the other decks waiting in the queue and picking the deck which is nearest to yours.
Vector may have a larger dimension than 3. But it's very computationally efficient to calculate.
EDIT: The dot product of 2 vectors is, to use a technical term, a measure of how pointy in the same direction they are.
For 2 vectors a and b,
a.b = (length of a) * (length of b) * pointiness in same direction multiplier.
Pointiness in the same direction multiplier is +1 if they are exactly pointing the same way. -1 if they are pointing opposite directions. And 0 if they are at right-angles. And various values between -1 and +1 for pointiness within all possible pointiness ranges. (Something to do with cosines and trigonometry goes here).
First of all, this is not a conspiracy thread. This is not to secretly complain about how the matchmaking system is rigged or whatever else, so, please, do not talk about it here.
I've been reading a lot on the forums about matchmaking, and I couldn't find any threads that really talk about specifically improving the matchmaking system. I've seen a few posts in other threads where people mention that there seems to be a secondary system in addition to your ranking that helps determine your next match-up, but I don't know if this is true or even provable outside of taking Blizzard's word about how the matchmaking works, and this is not the main point of the thread.
So, is there a better way to decide matches in Hearthstone that any of you have thought or heard of? This would be in addition to the Ranking System used below Legend, and in addition to the MMR of Legend and Casual. If you are going to use some form of statistic to help your argument, please make it easy to be understood for other people who may see it, and most importantly, it needs to be definitive, with some form of proof.
This is not to try to say we are better than Blizzard or anything of the sort, I am just curious about other people's ideas on how it could be improved, and once again, this is not a salt or conspiracy thread, so do not post about anything of the like.
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
Y'know, I honestly thought I'd get at least a few responses when I made this.
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
No I agree with you man..I feel honestly that everytime your about to hit winstreak that they match you with a counter deck type..I know there’s probably a better way to put it but I feel like it’s waaaaay to consistent for it to just be a coincidence..it’s happened with every class for me and it doesn’t matter what you do
For statistical porpoises, I'd like the matchmaking system to avoid rematching against the same opponent twice in a row if it possibly can without a long wait time because that knackers the assumptions of statistical tests that the opponent is drawn at random from the pool of all available suitable players. The reason rematches are common (especially at high ranks/late night early morning games) is because the pool of players near your rank is low and you both probably hit the Play button at about the same time after your previous match.
Rematches should be excluded as outliers from sample data since you are very likely to face the exact same deck -> the events are not independent of one another -> skews the data about distribution of classes played on ladder (for small sample sizes anyway. If you collect data from a lot of players, it becomes irrelevant).
That is all :P
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
Well the only thing I can think of off the top of my head would be an algorithm like the PageRank algorithm Google used as their first search algorithm to match decks of similar quality against each other.
You'd have to give a score for each card and combinations of cards used in the deck to come up with a rank for each players deck compared to current win rates then match decks of similar rating against each other. Vicious Syndicate has charts which show how common each card is in a particular deck archetype, this could be used as basis for a score, and the win rate for the archetype would be an additional factor on top.
https://en.wikipedia.org/wiki/PageRank
EDIT: You would use something like this to rate how similar a deck is to an existing archetype.
One does not simply walk into Mordor,
unless they want to be the best they can be.
PageRank style systems aren't about matching up counter decks, they are about matching decks of similar quality. Each deck would get a rating based on similarity to standard deck X which has quality Y and matching up to similarly rated decks. EDIT: You basically give each deck a rating and your deck is measured against similar decks by similarity to the reference deck. Search algorithms for the interwebz use this kind of thing to sort the results.
This could bring in a Millhouse Manastorm meta of course to throw the quality calculation off ;)
EDIT: Another problem with it would be that if there is a clearly superior deck then anyone playing it would often end up playing a mirror match. That could be amusing for a while I suppose.
This is a very interesting idea of how to improve matchmaking, but is it feasible at all? How hard/easy would it be to game the system to your needs to climb, and how difficult would it be to implement this system?
I'm still also interested in hearing other ideas of how to improve, so feel free to share any ideas you've had!
If you need a Wild deck that is fun, competitive, high skill, and low RNG, check out my Highlander Malygos Deck!
It would be feasible (just compare cards that are missing from the reference deck and adjust the score to compensate).
It can be gamed (put in deliberately bad cards and hope you don't draw them/never play them).
It's not difficult to implement if you know everyone's deck (like they do, it gets sent to the servers).
It's fast as well, which is why Google killed Yahoo as the best search engine very quickly. It basically involves doing a dot product of your decks score (which is a vector i.e. a bunch of numbers whose order is important, so multi-dimensional rather than a single number) against the other decks waiting in the queue and picking the deck which is nearest to yours.
Dot product of 2 vectors: (x,y,z).(x',y',z') = x*x' + y*y' + z*z'.
Vector may have a larger dimension than 3. But it's very computationally efficient to calculate.
EDIT: The dot product of 2 vectors is, to use a technical term, a measure of how pointy in the same direction they are.
For 2 vectors a and b,
a.b = (length of a) * (length of b) * pointiness in same direction multiplier.
Pointiness in the same direction multiplier is +1 if they are exactly pointing the same way. -1 if they are pointing opposite directions. And 0 if they are at right-angles. And various values between -1 and +1 for pointiness within all possible pointiness ranges. (Something to do with cosines and trigonometry goes here).