Subscribe
Special Developer Log #4 - Moving away from ELO and S13
Posted on 13 July 2016 - 02:09 AM
R3KoN wrote
Oh boy.
- Skill is not perfectly quantifiable via a single number. Rating systems yield approximates. All rating systems, Elo included, have a degree of inaccuracy. There is no true measure of skill, that's *exactly* why we develop rating systems.
- Differing playstyles and diverse skillsets don't cleanly fit into your "I am 16x better than _" world. See Dota2, LoL; players with the same rating can have incomparable skillsets.
- Ranked playerbase distribution data allows you to compare yourself to the community. If players in Master-tier are in the 99th or 100th percentile, you can make a good guess.
Unfortunately we can't just go off of percentile either. So far during season 12, about 300,000 players have played ranked Build. This puts everyone in the top 1000 in the top 0.33% of all Build UHC players. But Danteh's Elo is 675 higher than the 1,000th rated player (and 598 higher than the 500th). So clearly there is a world of difference even just between the players in the top 1000 (0.3%). With out Elo how can you ever tell whether the #1 player in (for example) horse has preformed 5x better than the #10 horse player or if he has merely done 1.5x as well?
Lists and percentiles are excellent for telling you who's better than who. But if you want to know how much better you need Elo.
Last edited on 13 July 2016 - 02:36 AM by Monasuko
SmellyPenguin wrote
BaneOfArthropods wrote...
LOL WTF?
When you get in a match and kill someone in a match you gain Elo. If he kills you, you lose Elo. The spread between winning and losing is always 32 Elo. In other words if you win you will be 32 Elo higher than if you lost.
If a player rated 2000 kills one rated 1400 he will gain 1 Elo, but if that 2k rated player dies instead then the 1400 player will gain a sweet 31 Elo. This tell us that the 2k player must win this fight 31 out of 32 times to keep up his Elo. So basic math tell us we can expect a 2000 rated player to beat a 1400 rated player 96.875% (yes, I know, that is way to many significant digits) of the time.
Unfortunately, there is no parallel with this new system. There is no way to tell if a Bronze 5 fighting a Gold I is a 1 in 20 underdog or a 1 in 50. It's ranking at its worst, numbers and medals for their own sake, with no clear measure of skill. And skill is all that matters.
Posted on 13 July 2016 - 03:36 AM
BaneOfArthropods wrote
SmellyPenguin wrote...
When you get in a match and kill someone in a match you gain Elo. If he kills you, you lose Elo. The spread between winning and losing is always 32 Elo. In other words if you win you will be 32 Elo higher than if you lost.
If a player rated 2000 kills one rated 1400 he will gain 1 Elo, but if that 2k rated player dies instead then the 1400 player will gain a sweet 31 Elo. This tell us that the 2k player must win this fight 31 out of 32 times to keep up his Elo. So basic math tell us we can expect a 2000 rated player to beat a 1400 rated player 96.875% (yes, I know, that is way to many significant digits) of the time.
Unfortunately, there is no parallel with this new system. There is no way to tell if a Bronze 5 fighting a Gold I is a 1 in 20 underdog or a 1 in 50. It's ranking at its worst, numbers and medals for their own sake, with no clear measure of skill. And skill is all that matters.
Your logic is seriously flawed dude, all I'm gonna say.
Posted on 13 July 2016 - 04:45 AM
SmellyPenguin wrote
BaneOfArthropods wrote...
Your logic is seriously flawed dude, all I'm gonna say.
Posted on 13 July 2016 - 07:18 AM
Vanthex wrote
GitRektSkrub wrote...
Bronze, Silver, Gold, Platinum, Diamond, Emerald
but friggin bronze silver and platinum arent a thing in mc…
and if u really wanna do it about metals like in olimpics or something, y diamond and emerald?
Posted on 13 July 2016 - 08:35 AM
Will the old, vertical kohi kb that we know and love be returning?And also nice, delayed/canceled hits (and maybe pot lag) will be gone. I like how this is looking.
good luck devs
Posted on 13 July 2016 - 08:38 AM
This seems awesome, great job Archy and rest of the devs :DPosted on 13 July 2016 - 09:22 AM
Not sure about this one, lets see how it goes….THIS IS THE BIGGEST COPY OF CLASH OF CLANS THO LOL
Posted on 13 July 2016 - 01:59 PM
" but they've been fixed in S13 since we use a multi server system that has no more then 50 people playing on at a time"Is this a problem if you're playing a non popular ladder? I mainly play NoDebuff, I'm confused. Could you explain it to me? Sorry
Posted on 13 July 2016 - 02:27 PM
Armoda wrote
" but they've been fixed in S13 since we use a multi server system that has no more then 50 people playing on at a time"
Is this a problem if you're playing a non popular ladder? I mainly play NoDebuff, I'm confused. Could you explain it to me? Sorry
Is this a problem if you're playing a non popular ladder? I mainly play NoDebuff, I'm confused. Could you explain it to me? Sorry
It's not like the actual situation where people queue up from a single server (which has 850 people at max), now it doesn't matter which server you queue up from, if you find an opponent it will tp you to a server.
Posted on 13 July 2016 - 02:38 PM
Saimo wrote
Armoda wrote...
It's not like the actual situation where people queue up from a single server (which has 850 people at max), now it doesn't matter which server you queue up from, if you find an opponent it will tp you to a server.
Posted on 13 July 2016 - 10:35 PM
@MonasukoYou are arguing over a bunch of silliness. ELO is supposed to define skill. MMR defines skill as well (but with better error calculations). I specifically modified the beta to show you how much "rating" you gain or lose so you can see yourself progress still. The only thing I can think of still adding is a % bar to show how close you are to ranking up or down. Otherwise there really is no point in providing further information. Based on our calculations getting Diamond/Masters will be very hard based on the # of players and their existing ratings. Instead of aiming for 2k or 2.5k you would aim for Plat 2 or Masters or something instead (I am just picking random names, these don't actually correlate).
Your arguments are based on completely psychological aspects that YOU as an individual cannot overcome, but not any actual proper logic for why ELO should be used instead of the new proposed system.
Posted on 14 July 2016 - 12:10 AM
MasterGberry wrote
@Monasuko
You are arguing over a bunch of silliness. ELO is supposed to define skill. MMR defines skill as well (but with better error calculations). I specifically modified the beta to show you how much "rating" you gain or lose so you can see yourself progress still. The only thing I can think of still adding is a % bar to show how close you are to ranking up or down. Otherwise there really is no point in providing further information. Based on our calculations getting Diamond/Masters will be very hard based on the # of players and their existing ratings. Instead of aiming for 2k or 2.5k you would aim for Plat 2 or Masters or something instead (I am just picking random names, these don't actually correlate).
Your arguments are based on completely psychological aspects that YOU as an individual cannot overcome, but not any actual proper logic for why ELO should be used instead of the new proposed system.
You are arguing over a bunch of silliness. ELO is supposed to define skill. MMR defines skill as well (but with better error calculations). I specifically modified the beta to show you how much "rating" you gain or lose so you can see yourself progress still. The only thing I can think of still adding is a % bar to show how close you are to ranking up or down. Otherwise there really is no point in providing further information. Based on our calculations getting Diamond/Masters will be very hard based on the # of players and their existing ratings. Instead of aiming for 2k or 2.5k you would aim for Plat 2 or Masters or something instead (I am just picking random names, these don't actually correlate).
Your arguments are based on completely psychological aspects that YOU as an individual cannot overcome, but not any actual proper logic for why ELO should be used instead of the new proposed system.
If an underdog/favorite scale could be added to the system, I would literally never complain about this again. Perhaps the program could tell you what it thinks your odds are when you enter a match: for example, it could say something like "Now in a match against Danteh, Rating Diamond I. You are a 1/15 underdog".
Am I making sense? Cause I can reword it if I'm not. This is pretty important to me.
Thank you for reading.
Posted on 14 July 2016 - 12:27 AM
I've always loved analytics. Especially in a sports context. I don't really GAF whether the Seahawks are better than the Patriots. What interests me is whether the Hawks have a 55% chance of winning against or a 65% of losing to the Pats, or whatever the number is that week. A power ranking listing aa 32 teams from #1 to #32 might be what 99% of fans are interested in. But as a numbers geek all I care about is the percentages. The skill gap if you will. You can't get percentages from a #1-#32 list, you need other data for that. All I really want is an Underdog/Favorite scale. It shouldn't be too hard to create, the data is all there already. I'll pay you for it. jsPosted on 14 July 2016 - 01:18 AM
BaneOfArthropods wrote
MasterGberry wrote...
If an underdog/favorite scale could be added to the system, I would literally never complain about this again. Perhaps the program could tell you what it thinks your odds are when you enter a match: for example, it could say something like "Now in a match against Danteh, Rating Diamond I. You are a 1/15 underdog".
Am I making sense? Cause I can reword it if I'm not. This is pretty important to me.
Thank you for reading.
Sure haven't had time to cover the whole thread. The skill gap between ranks is going to be a bell curve similar to how ELO already is. The bell curve will be more smooth. As per most games and ours as well based on initial calculations the % of players in each tier would be as follows:
Masters: ~0.00005%
Diamond: ~1%
Emerald: 4%
Gold: 14%
Silver: 42%
Bronze: 39%
(This is just the tiers)
We imagine the actual #s will be a bit different based on some factors such as the new matchmaking system etc.
To give some further graphics showing how the distribution of skill is currently vs the new system:
Red = ELO
Green = New MMR System
The left axis (Vertical) is the number of players.
The bottom axis (Horizontal) is the rating of the player.
As you can see the new system has a much better distribution of players unlike ELO which has a clusterfuck around the middle of the rating system.
The ratings are broken into equal distributions and we are going to limit matchmaking (I believe) to only allow 1 tier difference (e.g. Bronze V vs Silver I at worst case [might even be more strict on this]). If you are a tier difference than there is already a huge rating gap and you are considered a clear underdog. If you are a bronze V fighting a bronze III you are a slight underdog but not terribly (imagine 100 elo difference [not accurate/reflective of the real data]). For the most part you should be more concerned about where you are and your own ranking than the people's around you.
The amount of rating you gain/lose is dependent on several factors which are too complicated/confidential to get into. No point in showing who is the "underdog" and what not. You are either in it to win it or you aren't. It's that simple.
Hope this brings some clarity to the thread.
Last edited on 14 July 2016 - 01:24 AM by PizzqWithAQ
Love it takes out rage and makes me care even less about rating/elo but the names should be a bit more like Dirt, Coal, Iron, Gold, Diamond, Emerald Masters idk its not rly a minecraft thing and this is minecraft.Posted on 14 July 2016 - 02:10 AM
PizzqWithAQ wrote
Love it takes out rage and makes me care even less about rating/elo but the names should be a bit more like Dirt, Coal, Iron, Gold, Diamond, Emerald Masters idk its not rly a minecraft thing and this is minecraft.
YEAH GUYS I'M DIRT V RANK!
lol, sorry i just had to make fun of it
Posted on 14 July 2016 - 05:33 AM
MasterGberry wrote
Sure haven't had time to cover the whole thread. The skill gap between ranks is going to be a bell curve similar to how ELO already is. The bell curve will be more smooth. As per most games and ours as well based on initial calculations the % of players in each tier would be as follows:
Masters: ~0.00005%
Diamond: ~1%
Emerald: 4%
Gold: 14%
Silver: 42%
Bronze: 39%
(This is just the tiers)
We imagine the actual #s will be a bit different based on some factors such as the new matchmaking system etc.
To give some further graphics showing how the distribution of skill is currently vs the new system:
Red = ELO
Green = New MMR System
The left axis (Vertical) is the number of players.
The bottom axis (Horizontal) is the rating of the player.
As you can see the new system has a much better distribution of players unlike ELO which has a clusterfuck around the middle of the rating system.
The ratings are broken into equal distributions and we are going to limit matchmaking (I believe) to only allow 1 tier difference (e.g. Bronze V vs Silver I at worst case [might even be more strict on this]). If you are a tier difference than there is already a huge rating gap and you are considered a clear underdog. If you are a bronze V fighting a bronze III you are a slight underdog but not terribly (imagine 100 elo difference [not accurate/reflective of the real data]). For the most part you should be more concerned about where you are and your own ranking than the people's around you.
The amount of rating you gain/lose is dependent on several factors which are too complicated/confidential to get into. No point in showing who is the "underdog" and what not. You are either in it to win it or you aren't. It's that simple.
Hope this brings some clarity to the thread.
Maybe I'll figure it out on my own someday.
Posted on 14 July 2016 - 06:09 AM
MasterGberry wrote
BaneOfArthropods wrote...
Sure haven't had time to cover the whole thread. The skill gap between ranks is going to be a bell curve similar to how ELO already is. The bell curve will be more smooth. As per most games and ours as well based on initial calculations the % of players in each tier would be as follows:
Masters: ~0.00005%
Diamond: ~1%
Emerald: 4%
Gold: 14%
Silver: 42%
Bronze: 39%
(This is just the tiers)
We imagine the actual #s will be a bit different based on some factors such as the new matchmaking system etc.
To give some further graphics showing how the distribution of skill is currently vs the new system:
Red = ELO
Green = New MMR System
The left axis (Vertical) is the number of players.
The bottom axis (Horizontal) is the rating of the player.
As you can see the new system has a much better distribution of players unlike ELO which has a clusterfuck around the middle of the rating system.
The ratings are broken into equal distributions and we are going to limit matchmaking (I believe) to only allow 1 tier difference (e.g. Bronze V vs Silver I at worst case [might even be more strict on this]). If you are a tier difference than there is already a huge rating gap and you are considered a clear underdog. If you are a bronze V fighting a bronze III you are a slight underdog but not terribly (imagine 100 elo difference [not accurate/reflective of the real data]). For the most part you should be more concerned about where you are and your own ranking than the people's around you.
The amount of rating you gain/lose is dependent on several factors which are too complicated/confidential to get into. No point in showing who is the "underdog" and what not. You are either in it to win it or you aren't. It's that simple.
Hope this brings some clarity to the thread.
Last edited on 14 July 2016 - 10:52 AM by PizzqWithAQ
MasterGberry wrote
PizzqWithAQ wrote...
YEAH GUYS I'M DIRT V RANK!
lol, sorry i just had to make fun of it
-.- +1 unfollow on twitter