I understand that, my point was that a new #1 game would have double the impact, since it would change the values of 2 of the games in the table, not just one.
How about use the median game if there are less than say, 5-8 games in a table?
Do we dare to try to fix QScore?
- Thread starter Denniz
- Start date
I understand that, my point was that a new #1 game would have double the impact, since it would change the values of 2 of the games in the table, not just one.
How about use the median game if there are less than say, 5-8 games in a table?
I picked those because they were tables with many submissions that still had outliers - meaning 50%BestTurn likely raises the average which isn't right - since there's enough submissions in those tables, the relationship between table average and Qscore of the actual games is good as-is.
I think it's too complicated to selectively apply the phantom only to tables without "enough" submissions. But since the phantom has dramatically less effect on the table average in heavily populated tables, it won't affect Qscores very much, so I don't see a need to exclude Gauntlet tables anyway.
AMiraculix - I like WastinTime's BestTurn+50%AvgTurn. I'm not sure I understand why you say it doesn't work in sparsely populated tables. What's your concern?
I believe you are confusing two different proposals that are completely unrelated (although they might address similar problems):
1) Adjustable phantom date
I agree that MinTurn+50%MinTurn and MinTurn+50%AvgTurn would have roughly the same effect in most tables, and I have no problem with applying either formula in large tables. My comment on Gauntlet tables had nothing to do with this proposal. The only reservation against the latter formula is that I believe avgTurn is an unreliable parameter in tables with few entries. I believe this is true whether one uses avgTurn to define the phantom date or to define the BaseQScore curve. That brings me to the second proposal.
2) Redefining the 50%-point of the BaseQScore curve (post 197)
Instead of 50%date = avgDate, use 50%turn = minTurn + X% (simple version). The purpose is to make the BaseQScore curve independent of all table entries except the #1 game. My comment on using the Gauntlet tables was only related to how to estimate X. Once X is estimated, I have no objections to applying the formula (simple or advanced version) to all tables, including Gauntlet tables.
PS! I am still waiting for someone to explain why this proposal would have problems in extreme cases. (Assuming we find a good estimate for X.)
To change the subject slightly, since I don't really have the type and depth of knowledge about HOF scoring and such as the others debating tis point.
Re: Balancing the EQM.
Since scoring penalties will not bother people who are just checking off boxes anyway, and since it may not have its intended effect by filtering out players who are 'unbalanced', why not give them the red unbalanced background that Denniz mentioned, and then filter them not off of the tables, but to the bottom of the rankings for each level of EQM. This way they are able to check off boxes if they wish, but they aren't really included in the rankings, as the 'worst' ranking balanced player will outrank them. When they get balanced again, they get reinserted into their proper spot in the rankings.
Re: Balancing the EQM.
Since scoring penalties will not bother people who are just checking off boxes anyway, and since it may not have its intended effect by filtering out players who are 'unbalanced', why not give them the red unbalanced background that Denniz mentioned, and then filter them not off of the tables, but to the bottom of the rankings for each level of EQM. This way they are able to check off boxes if they wish, but they aren't really included in the rankings, as the 'worst' ranking balanced player will outrank them. When they get balanced again, they get reinserted into their proper spot in the rankings.
RobertTheBruce
King
Why not set a minimum number of turns for the average in the Qscore calculation?
The main sticking point in the Qscore calculation seems to be sparse tables where 'good' games get a very low Qscore because the variance of the distribution is very small.
For example, a table with a finish on turn 232. There really isn't much information about whether this is a remarkable, average or poor game. It could be a great diplomatic victory or a poor conquest. A second player adds a finish on turn 230 and the Qscore of the first game goes from 100 to 10. This is pretty harsh for a game that seems to be only slightly worse. If the second game finished on turn 180, the large drop in Qscore better reflects the new information about the relative quality of the first game.
You can try a minimum deviation (distance from best to average) of say (6 -# of games in the deviation calculation)*years per turn rate of best finish. This will not alter the Qscores in tables with many entries and the arbitrary scale factor can be tested to produce something reasonable depending on speed or other factors. This is a more direct method of preventing small variances compared to adding a phantom date.
If a new entry is significantly better than the old #1, the Qscore of the old #1 will drop significantly (consider the straw man when the T232 finish is beaten by a T180 finish). If it is only marginally better, the Qscore will only drop marginally. Adding a phantom finish doesn't reflect this new information in most cases. If an additional game with a finish on T660 is added to the table, the Qscore of the T232 finish will be similar after the second game is added whether it is a finish at T230 (the first game still looks good) or at T180 (the first game now looks pretty poor compared to the best game). With the original system, the Qscore drops dramatically in both cases; with the phantom game, it doesn't drop dramatically in either case. If 6 or 7 good players can set very similar dates, a small variance is probably a good description of the distribution and games slightly later get a reasonable low Qscore.
This still leaves the problem with 'poor' games added to a table of 'good' games. The ratio of good to bad games in a table has a strong effect on the Qscore of marginal 'good' games. I don't know if its possible to simply resolve that issue without knowing what a 'average' finish date is for each table independent of the entries. For example, one can hide games in a table with a large number of late finish dates due to a gauntlet with restrictive conditions. Its possible to game the formula to get good Qscores for difficult leaders or map scripts if you want a high Qscore for the QM or EQM by stuffing them into the tables where gauntlets skew the distribution. No one seems to complain about getting high Qscores for mediocre games.
The scale factors which add to or multiply the Qscore are another can of worms I won't open.
The main sticking point in the Qscore calculation seems to be sparse tables where 'good' games get a very low Qscore because the variance of the distribution is very small.
For example, a table with a finish on turn 232. There really isn't much information about whether this is a remarkable, average or poor game. It could be a great diplomatic victory or a poor conquest. A second player adds a finish on turn 230 and the Qscore of the first game goes from 100 to 10. This is pretty harsh for a game that seems to be only slightly worse. If the second game finished on turn 180, the large drop in Qscore better reflects the new information about the relative quality of the first game.
You can try a minimum deviation (distance from best to average) of say (6 -# of games in the deviation calculation)*years per turn rate of best finish. This will not alter the Qscores in tables with many entries and the arbitrary scale factor can be tested to produce something reasonable depending on speed or other factors. This is a more direct method of preventing small variances compared to adding a phantom date.
If a new entry is significantly better than the old #1, the Qscore of the old #1 will drop significantly (consider the straw man when the T232 finish is beaten by a T180 finish). If it is only marginally better, the Qscore will only drop marginally. Adding a phantom finish doesn't reflect this new information in most cases. If an additional game with a finish on T660 is added to the table, the Qscore of the T232 finish will be similar after the second game is added whether it is a finish at T230 (the first game still looks good) or at T180 (the first game now looks pretty poor compared to the best game). With the original system, the Qscore drops dramatically in both cases; with the phantom game, it doesn't drop dramatically in either case. If 6 or 7 good players can set very similar dates, a small variance is probably a good description of the distribution and games slightly later get a reasonable low Qscore.
This still leaves the problem with 'poor' games added to a table of 'good' games. The ratio of good to bad games in a table has a strong effect on the Qscore of marginal 'good' games. I don't know if its possible to simply resolve that issue without knowing what a 'average' finish date is for each table independent of the entries. For example, one can hide games in a table with a large number of late finish dates due to a gauntlet with restrictive conditions. Its possible to game the formula to get good Qscores for difficult leaders or map scripts if you want a high Qscore for the QM or EQM by stuffing them into the tables where gauntlets skew the distribution. No one seems to complain about getting high Qscores for mediocre games.
The scale factors which add to or multiply the Qscore are another can of worms I won't open.
1) I would remove the average calculation altogether for tables with less than 3 entries (I mean 3 different players, not 3 times the same). Just use the last date as "average". It means so little for 2 entries it's just not fun.
2) I think phantom date is only a solution for very hard settings, where only a handful of players can get a win, and thess wins should be worth something, even if late. Maybe populating the large/huge normal and quick deity tables with 1800AD phantoms would be good enough?
2) I think phantom date is only a solution for very hard settings, where only a handful of players can get a win, and thess wins should be worth something, even if late. Maybe populating the large/huge normal and quick deity tables with 1800AD phantoms would be good enough?
Dracandross
Warlord
Hmm caberts suggestion is also good and quite simple to understand.
Any solution that negates itself when there are enough games is good...
That would at least increase submissions for hard settings but Id say put it to 5 instead of 3 so there are enough submissions afterwards...
-D
Any solution that negates itself when there are enough games is good...
That would at least increase submissions for hard settings but Id say put it to 5 instead of 3 so there are enough submissions afterwards...
-D
CliftonBazaar
Prince
To make the EQM REALLY hard why not allow games that are only in the HOF tables - I.E the top ten.
If a game drops out of the top ten then it no longer counts towards the EQM.
This idea would stop anyone who is trying to 'tick the boxes' in their tracks.
which leads me to the next subject (read on)
I'd like to point out that I'm one of those who are 'ticking the boxes', but I'm also trying to get high scores (I like seeing them in the top ten).
So what do I do after the boxes are all ticked? I try to improve them. When the next update takes place I will be in the EQM for settler - Warlord. So I'm now trying to improve them all in order to be a 'Noble EQM' (has a certain ring to it
)
And why are most of my games on small maps? Because this computer has problems after 1800AD and I need to finish the games before that.
If a game drops out of the top ten then it no longer counts towards the EQM.
This idea would stop anyone who is trying to 'tick the boxes' in their tracks.
which leads me to the next subject (read on)
I'd like to point out that I'm one of those who are 'ticking the boxes', but I'm also trying to get high scores (I like seeing them in the top ten).
So what do I do after the boxes are all ticked? I try to improve them. When the next update takes place I will be in the EQM for settler - Warlord. So I'm now trying to improve them all in order to be a 'Noble EQM' (has a certain ring to it
)And why are most of my games on small maps? Because this computer has problems after 1800AD and I need to finish the games before that.
I have been goofing off again.
So far the best idea is having multiple phantom dates based on different "best date" ranges. So far I haven't thought of a scientific method of picking the dates or ranges but it probably doesn't matter too much as long as they are closer together than a single phantom date would be.
The other thing that I have been think about it the idea a coupe of you have had with starting with the best game at 100 and subtracting 1 for each turn. At first I didn't think much of the idea but it is starting to grow on me. Can anybody point out the potential flaws?
I was thinking of creating an alternate versions of the EQM and Game Info pages with one of these methods so we can check the effects.
So far the best idea is having multiple phantom dates based on different "best date" ranges. So far I haven't thought of a scientific method of picking the dates or ranges but it probably doesn't matter too much as long as they are closer together than a single phantom date would be.
The other thing that I have been think about it the idea a coupe of you have had with starting with the best game at 100 and subtracting 1 for each turn. At first I didn't think much of the idea but it is starting to grow on me. Can anybody point out the potential flaws?
I was thinking of creating an alternate versions of the EQM and Game Info pages with one of these methods so we can check the effects.
CliftonBazaar
Prince
Early Inca wins will completely destroy all other games.
For example in Immortal, Standard size, Domination victory the earliest is 510BC while the second spot is 732AD - this would mean every other game, apart from the first, would get a score of zero. I realize that this isn't an Incan win (Hatty) but the earliest wins are usually Inca.
Having a look at the Duel maps (Domination) nearly all the early entries (on the high difficulty levels) have a UU very early in the game - Inca, Hatty and Cyrus. If you were playing any other leader then you are at a severe dis-advantage.
Great idea, then we can all see the facts before we comment - at the moment we are all discussing theories.
Harbourboy
Deity
What is wrong with ticking the boxes? That idea would prevent me from ever getting any EQM points at all, so I would be firmly against it. EQM and HoF tables should not be related that strongly. EQM should be for people who have 'crossed the finish line' in a wide variety of races. QScore and HoF Tables should be like the Guiness Book of Records or a Medal Table - to reward the very best performances in each event.
I really don't like the idea of dropping one point per turn, it is very unbalanced with the different speeds.
I do like the idea of having a test EQM page with some of these ideas played out so we can see what the real effects are.
I do like the idea of having a test EQM page with some of these ideas played out so we can see what the real effects are.
CliftonBazaar
Prince
Nothing is wrong with 'ticking the boxes', as I said I do it myself, but reading this whole thread I believe that the majority of people here believe that this is against the spirit of the EQM.
Harbourboy
Deity
There's a difference between ticking the boxes to reach the EQM objective, and ticking the boxes and getting a high score. I'm all for the QScore rewarding great performances, but EQM should just reward entering the race and getting over the line, no matter how slowly or clumsily.
I was confused - all clear now. I don't have a problem with either proposal. I think either one will handle tables with few entries better than the current method.
With respect to your second proposal, it would not work for RoA games. Obviously if there is any decent ancient start game in a table, MinTurn+50% of whatever is not going to be sufficient to get a Future start more than a handful of Qscore points. Of course what we have now doesn't work for RoA either - right now you need an empty table, an oddball table with a well-below-average average date or a Time victory to get a decent RoA score in the mid-later eras. I think to be meaningful the RoA games would have to be scored separately or it should be dropped alltogether (which I'm not in favor of since I enjoy the different gameplay in later era starts).
I believe this is the right direction. Some comments:
1) The BaseQScore would be independent of all table entries except the #1 game. This is a good thing for tables with few entries.
2) If we adopt this method or a similar metod, a phantom date would not be needed (nor will it have any effect).
3) One disadvantage is that every game finishing more than 100 turns after the #1 game would get 0 points. This is why I believe the 50%turn formula of post 197 would work better because it scales with minTurn.
Following is a short comparison of three different methods for assigning BaseQScores. The methods are compared with respect to 1) The curve to use abd 2) The 50%point (defined as the turn/date that results in 50 BaseQScore points).
Current method:
- curve: nonlinear (BaseQScore curve)
- 50%point: averageDate
One point-per-turn method:
- curve: linear
- 50%point: minTurn + 50
50%turn method
- curve: nonlinear (BaseQScore curve)
- 50%point: minTurn + X%
I believe the 50%turn method is better for the following reasons:
1) It is independent of all table entries except the #1 game.
2) The 50%point is scaled with minTurn.
3) It keeps the current BaseQScore curve unchanged.
Finally, the advanced method of post 197 could be considered a mix of the current method and the 50%turn method.
I have been playing with a new QScore formula based on the min date that scales by speed. It is a variation on the -1 point per turn idea. The big difference is that it uses the number of turns in the game.
You have to finish within 25% of the total game length of the best date to score. At this point I am using 25% as it feel better then some of the others.
Here is a formula I have been playing with:
QScore = 100 - (100 * (entry_turn - #1's_turn) / ((#1's_total_turns * 0.25) * (entry_total_turns / #1's_total_turns ))
Using the FiveAces 1310AD small, deity, diplo, quick entry as an example:
entry_turn = 134
entry_total_turns = 320
#1's_turn = 122
#1's_total_turns = 320
QScore = 100 - (100 * (134 - 122) / 320 * 0.25) * (320 / 320))
QScore = 100 - (100 * (12 / 80 * 1)
QScore = 100 - (100 * 0.15)
QScore = 100 - 15
QScore = 85
I have a query page where you can look at the effects (login before click link): QScore query
Note: the basic ratio of this formula might also be used to create a phantom date if this doesn't work out. I like this as it addresses both issues with the QScore.
___________________________
BTW, I remove the adjustors from the EQM page QScores. It is still the same QScore formula as before. Not the one I am testing.
___________________________
Also, one of the things I am going to look at, is use only games eligible for the EQM for purposes of selecting best game and calculating EQM QScore. I will integrate the test page at some point so people can see how each qscore is calculated.
You have to finish within 25% of the total game length of the best date to score. At this point I am using 25% as it feel better then some of the others.
Here is a formula I have been playing with:
QScore = 100 - (100 * (entry_turn - #1's_turn) / ((#1's_total_turns * 0.25) * (entry_total_turns / #1's_total_turns ))
Using the FiveAces 1310AD small, deity, diplo, quick entry as an example:
entry_turn = 134
entry_total_turns = 320
#1's_turn = 122
#1's_total_turns = 320
QScore = 100 - (100 * (134 - 122) / 320 * 0.25) * (320 / 320))
QScore = 100 - (100 * (12 / 80 * 1)
QScore = 100 - (100 * 0.15)
QScore = 100 - 15
QScore = 85
I have a query page where you can look at the effects (login before click link): QScore query
Note: the basic ratio of this formula might also be used to create a phantom date if this doesn't work out. I like this as it addresses both issues with the QScore.
___________________________
BTW, I remove the adjustors from the EQM page QScores. It is still the same QScore formula as before. Not the one I am testing.
___________________________
Also, one of the things I am going to look at, is use only games eligible for the EQM for purposes of selecting best game and calculating EQM QScore. I will integrate the test page at some point so people can see how each qscore is calculated.
Harbourboy
Deity
So I will get almost no QScores then? I am hardly ever within 25% of the best game, especially on the popular tables. This will discourage me from bothering to submit in any but the most obscure tables.
We are talking about finishing within 80 turns of the best game on quick speed.
25% turns by speed / division
Code:
[B][U]V&W[/U] [U]BTS[/U][/B]
Quick 80 82.5
Normal 115 125
Epic 165 187.5
Marathon 300 375
The idea is to recognize/reward better play. The results should show up in the tables. Right now it doesn't work out that way most of the time. The average date problems make QScore almost irrelevant right now.
If the new EQScore ends up being more meaningful than the old QScore wouldn't that give you incentive to learn to achieve better dates? I know people are always ready to help out here in the HOF Forum.
Similar threads
- Locked
