Forum > Goal Line Blitz 2 > Ladder Tracking
Biggest problem we have is we are dealing with a big corrective period for most teams because we had a huge glut of teams reset/quit.
People have suggested resetting ELO to a base rate for each ranking spot each season, but not sure if that works for the reasons for using ELO anyways.
People have suggested resetting ELO to a base rate for each ranking spot each season, but not sure if that works for the reasons for using ELO anyways.
I don't like that as much Gal..
but I could see supporting an idea like "only the last 60 games count and anything older drops off" -- that might be a nice balance
but I could see supporting an idea like "only the last 60 games count and anything older drops off" -- that might be a nice balance
Parab00n
offline
offline
Originally posted by TxSteve
I don't like that as much Gal..
but I could see supporting an idea like "only the last 60 games count and anything older drops off" -- that might be a nice balance
To do my best Romo impression, "it doesn't work like that". Drastically increasing the K-value, resetting the ELO, or reducing each teams ELO on a sliding scale(i.e. every team loses 10% of total ELO at the end of the season) I believe are the only options to get us where we want.
I don't like that as much Gal..
but I could see supporting an idea like "only the last 60 games count and anything older drops off" -- that might be a nice balance
To do my best Romo impression, "it doesn't work like that". Drastically increasing the K-value, resetting the ELO, or reducing each teams ELO on a sliding scale(i.e. every team loses 10% of total ELO at the end of the season) I believe are the only options to get us where we want.
genuine question....why not?
I'll just use a 10 game scale.
Say the ELO result of my last 10 games is:
+4
+2
+1
+9
-7
-4
+2
+3
-4
+7
That gives me a total ELO of 13
Now my next game gets me +5 -- but my +7 game drops off - resulting in a 11 ELO - so I won my most recent ladder game but still lost ELO because the win isn't as high quality.
I'll just use a 10 game scale.
Say the ELO result of my last 10 games is:
+4
+2
+1
+9
-7
-4
+2
+3
-4
+7
That gives me a total ELO of 13
Now my next game gets me +5 -- but my +7 game drops off - resulting in a 11 ELO - so I won my most recent ladder game but still lost ELO because the win isn't as high quality.
Xars
offline
offline
Their argument is that ELO doesn't use "old" games. It's just one number.
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
Edited by Xars on Apr 14, 2015 08:35:19
Edited by Xars on Apr 14, 2015 08:34:39
Parab00n
offline
offline
Originally posted by Xars
Their argument is that ELO doesn't use "old" games. It's just one number.
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
But it does matter, in that scenario Team A has 900 more games in which he gained or lost ELO.
Their argument is that ELO doesn't use "old" games. It's just one number.
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
But it does matter, in that scenario Team A has 900 more games in which he gained or lost ELO.
peeti
offline
offline
Originally posted by TxSteve
genuine question....why not?
I'll just use a 10 game scale.
Say the ELO result of my last 10 games is:
+4
+2
+1
+9
-7
-4
+2
+3
-4
+7
That gives me a total ELO of 13
Now my next game gets me +5 -- but my +7 game drops off - resulting in a 11 ELO - so I won my most recent ladder game but still lost ELO because the win isn't as high quality.
and you dont see why this is stupid? Glad elo downt work that way anyway
genuine question....why not?
I'll just use a 10 game scale.
Say the ELO result of my last 10 games is:
+4
+2
+1
+9
-7
-4
+2
+3
-4
+7
That gives me a total ELO of 13
Now my next game gets me +5 -- but my +7 game drops off - resulting in a 11 ELO - so I won my most recent ladder game but still lost ELO because the win isn't as high quality.
and you dont see why this is stupid? Glad elo downt work that way anyway
Xars
offline
offline
Originally posted by Xars
Their argument is that ELO doesn't use "old" games. It's just one number.
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
Originally posted by Parab00n
But it does matter, in that scenario Team A has 900 more games in which he gained or lost ELO.
Well yes and no.
If we both take Calculus and I've taken 100 practice tests at an average score of 93 and you've taken 10 practice tests at an average score of 85, the number of tests shouldn't matter in terms of "prediction".
And that's an ELO system. It's simply a prediction model that alters values depending on how different the result was from the prediction.
The problem in GLB2 is that the "INPUT" isn't the same is it is in Chess/whatever. Teams change season to season. Players leave, there are Chemistry hits, Owners go inactive, etc.
So when the old INPUT isn't similar to the new INPUT, then the ELO score is false. Could be a little false or a lot false.
It's why I recommended changing the ELO score of a team when their was high roster turnover. (Thread I got torched on.) But perhaps people are now realizing I was right.
Example: The Stunners have the highest ELO in the game. Every time they win, even though they are predicted to win, they gain a small (probably very small) amount of ELO.
What happens if TxSteve goes inactive and doesn't reset the Stunners next season? They go CPU with players at 50 chem.
Is that team the same as the one that created the ELO? No. The "new" team inherits the "old" teams ELO.
The INPUT driving the ELO score is drastically different. Yes it eventually self-corrects but not enough in a 30 game season.
However, if I play Chess I'll get an ELO score. Every time I play a new game of Chess, my ELO will change. However, the INPUT of that ELO (me) is constant.
That's not the case in GLB2 and it's why ELO needs to be radically adjusted in some way.
I still think my way of using Chemistry values to adjust ELO makes more sense than penalizing player performance on the field, but in order for that idea to get adopted all of you need to catch up to me.
Their argument is that ELO doesn't use "old" games. It's just one number.
Team A has ELO of X
Team B has ELO of Y
It shouldn't matter if Team A has played 1000 games and Team B only 100.
The rating is the rating.
http://en.wikipedia.org/wiki/Elo_rating_system
Originally posted by Parab00n
But it does matter, in that scenario Team A has 900 more games in which he gained or lost ELO.
Well yes and no.
If we both take Calculus and I've taken 100 practice tests at an average score of 93 and you've taken 10 practice tests at an average score of 85, the number of tests shouldn't matter in terms of "prediction".
And that's an ELO system. It's simply a prediction model that alters values depending on how different the result was from the prediction.
The problem in GLB2 is that the "INPUT" isn't the same is it is in Chess/whatever. Teams change season to season. Players leave, there are Chemistry hits, Owners go inactive, etc.
So when the old INPUT isn't similar to the new INPUT, then the ELO score is false. Could be a little false or a lot false.
It's why I recommended changing the ELO score of a team when their was high roster turnover. (Thread I got torched on.) But perhaps people are now realizing I was right.
Example: The Stunners have the highest ELO in the game. Every time they win, even though they are predicted to win, they gain a small (probably very small) amount of ELO.
What happens if TxSteve goes inactive and doesn't reset the Stunners next season? They go CPU with players at 50 chem.
Is that team the same as the one that created the ELO? No. The "new" team inherits the "old" teams ELO.
The INPUT driving the ELO score is drastically different. Yes it eventually self-corrects but not enough in a 30 game season.
However, if I play Chess I'll get an ELO score. Every time I play a new game of Chess, my ELO will change. However, the INPUT of that ELO (me) is constant.
That's not the case in GLB2 and it's why ELO needs to be radically adjusted in some way.
I still think my way of using Chemistry values to adjust ELO makes more sense than penalizing player performance on the field, but in order for that idea to get adopted all of you need to catch up to me.
Edited by Xars on Apr 14, 2015 09:07:05
Edited by Xars on Apr 14, 2015 09:06:26
Also realize how ELO is self-correcting. It's not based on your ladder rank it's based on your ELO value. This is a big asset to the system that was chosen here.
So (hypothetically) let's say Team A, rank 3 has an ELO of 1000. They're older than Team B, rank 4 who only has an ELO of 800 and is only rank 4 due to a bunch of teams retiring. If Team B beats Team A they are going to get a sizable increase in ELO. That's a huge upset, the same perceptive way as if that previous team was rank 15 beating rank 3. The rank 4 team beating rank 3 team isn't stuck being a 'small' upset. That ELO difference, even in the situation of one team being older, helps the ELO values correct themselves. This is one of the big flaws we have now with not being able to see the ELO values. We can't see the big upsets and how a team was rewarded.
So in reality, if the k-value was appropriately adjusted and we simultaneously could see the ELO values, we'd have no problem with the system...assuming we don't get consistent crashes of 10+ teams from the top 20 each season, lol.
Team 4 who is a season younger or w/e could have their value corrected faster.
What won't happen is people jumping from rank 40 to rank 1-5 in 10 games. I seriously think there are people who want/expect that here. That's a crazy system where matchups would fuck you over from being appropriately ranked.
So (hypothetically) let's say Team A, rank 3 has an ELO of 1000. They're older than Team B, rank 4 who only has an ELO of 800 and is only rank 4 due to a bunch of teams retiring. If Team B beats Team A they are going to get a sizable increase in ELO. That's a huge upset, the same perceptive way as if that previous team was rank 15 beating rank 3. The rank 4 team beating rank 3 team isn't stuck being a 'small' upset. That ELO difference, even in the situation of one team being older, helps the ELO values correct themselves. This is one of the big flaws we have now with not being able to see the ELO values. We can't see the big upsets and how a team was rewarded.
So in reality, if the k-value was appropriately adjusted and we simultaneously could see the ELO values, we'd have no problem with the system...assuming we don't get consistent crashes of 10+ teams from the top 20 each season, lol.
Team 4 who is a season younger or w/e could have their value corrected faster.
What won't happen is people jumping from rank 40 to rank 1-5 in 10 games. I seriously think there are people who want/expect that here. That's a crazy system where matchups would fuck you over from being appropriately ranked.
Galithor
offline
offline
Originally posted by InRomoWeTrust
What won't happen is people jumping from rank 40 to rank 1-5 in 10 games. I seriously think there are people who want/expect that here. That's a crazy system where matchups would fuck you over from being appropriately ranked.
or, it could more like the real football system GLB2 is most closely modeled after (college rankings), where that's exactly what should happen.
That we let previous season results impact the current season rankings soo much is entirely the problem with the way it's being done. That's why we can't shed CPU/gutted teams. It's why really good teams ranked in the teens have no prayer of attaining #1.
If we started with a clean slate like Sagarin or RPI every season, every season would look alot better and make a hell of alot more sense.
Elo works for mighty brawlers because of the never ending nature of the wins/losses. GLB2 has definitive start and end points every 30 games where the nature of teams get signficantly altered. ELO does not make sense for that structure.
What won't happen is people jumping from rank 40 to rank 1-5 in 10 games. I seriously think there are people who want/expect that here. That's a crazy system where matchups would fuck you over from being appropriately ranked.
or, it could more like the real football system GLB2 is most closely modeled after (college rankings), where that's exactly what should happen.
That we let previous season results impact the current season rankings soo much is entirely the problem with the way it's being done. That's why we can't shed CPU/gutted teams. It's why really good teams ranked in the teens have no prayer of attaining #1.
If we started with a clean slate like Sagarin or RPI every season, every season would look alot better and make a hell of alot more sense.
Elo works for mighty brawlers because of the never ending nature of the wins/losses. GLB2 has definitive start and end points every 30 games where the nature of teams get signficantly altered. ELO does not make sense for that structure.
I disagree and trying to compare ELO to college football is a bad bad idea as they're so functionally different.
Dominant teams like the Stunners or Queen City before them or Doinitin Djibouti for them, etc. etc. shouldn't be yanked from top spots because of some individuals desire for a volatile system. We don't have the scheduling system for something so volatile. It's easy to get a bunch of waste games in a row and in a volatile system that'd tank your rank.
The season would NOT look alot better having a team like the Stunners at ladder rank 10 or w/e they'd end up with due to schedule fluctuations.
In an ideal ELO system, the k-value is set so that the rankings correct themselves in a timely manner (~15-20 games?). That's really all we're looking for (and being able to see the damn ELO scores
).
Dominant teams like the Stunners or Queen City before them or Doinitin Djibouti for them, etc. etc. shouldn't be yanked from top spots because of some individuals desire for a volatile system. We don't have the scheduling system for something so volatile. It's easy to get a bunch of waste games in a row and in a volatile system that'd tank your rank.
The season would NOT look alot better having a team like the Stunners at ladder rank 10 or w/e they'd end up with due to schedule fluctuations.
In an ideal ELO system, the k-value is set so that the rankings correct themselves in a timely manner (~15-20 games?). That's really all we're looking for (and being able to see the damn ELO scores
).Edited by InRomoWeTrust on Apr 14, 2015 10:13:47
I disagree.
No matter past records -- if the stunners drop their next 3-4 games...while Harrisonburg wins their next 3-4 -- It is my opinion that the game would be better and more fun if we swapped places (or if the stunners even dropped further if MMA keeps winning).
Ladder changes are fun.
Meaningful games are fun.
Games and results mattering is fun.
We are approaching a time (since stunners are resetting) when ladder games are going to mean absolutely SQUAT. I might even be there...in a spot were it really does not matter if the stunners win or lose (I'm assuming we have several auto-wins on our ladder schedule from CPU/etc). That isn't fun - and actually promotes exactly what you did which made the game suck --- throwing in a default game plan and just letting it fly since I'd still win 1/2 or more of my games and hang onto #1.
Same with Hburg or MMA or Air Raid. We are approaching a point in the season (probably already there) where the best they can probably hope for is to possibly scoot up a single spot -- but mostly just hang on through the season to get their 2nd/3rd/4th trophy. Nothing fun about that either.
No matter past records -- if the stunners drop their next 3-4 games...while Harrisonburg wins their next 3-4 -- It is my opinion that the game would be better and more fun if we swapped places (or if the stunners even dropped further if MMA keeps winning).
Ladder changes are fun.
Meaningful games are fun.
Games and results mattering is fun.
We are approaching a time (since stunners are resetting) when ladder games are going to mean absolutely SQUAT. I might even be there...in a spot were it really does not matter if the stunners win or lose (I'm assuming we have several auto-wins on our ladder schedule from CPU/etc). That isn't fun - and actually promotes exactly what you did which made the game suck --- throwing in a default game plan and just letting it fly since I'd still win 1/2 or more of my games and hang onto #1.
Same with Hburg or MMA or Air Raid. We are approaching a point in the season (probably already there) where the best they can probably hope for is to possibly scoot up a single spot -- but mostly just hang on through the season to get their 2nd/3rd/4th trophy. Nothing fun about that either.
Originally posted by TxSteve
No matter past records -- if the stunners drop their next 3-4 games...while Harrisonburg wins their next 3-4 -- It is my opinion that the game would be better and more fun if we swapped places (or if the stunners even dropped further if MMA keeps winning).
That's EXACTLY what I was arguing for (adjusting the k-value takes care of that).
What I was arguing against is a system where Minnesota goes 5-0 and Queen City goes 5-0. Queen City had a tougher 5 games so Queen City is ranked above Minnesota. Completely ignoring that Minnesota has won it's last 30 games (or w/e you want for the hypothetical)
No matter past records -- if the stunners drop their next 3-4 games...while Harrisonburg wins their next 3-4 -- It is my opinion that the game would be better and more fun if we swapped places (or if the stunners even dropped further if MMA keeps winning).
That's EXACTLY what I was arguing for (adjusting the k-value takes care of that).
What I was arguing against is a system where Minnesota goes 5-0 and Queen City goes 5-0. Queen City had a tougher 5 games so Queen City is ranked above Minnesota. Completely ignoring that Minnesota has won it's last 30 games (or w/e you want for the hypothetical
)You are not logged in. Please log in if you want to post a reply.
