Conquer Club

chipv wrote:

BaldAdonis wrote:

Incandenza wrote:Yeah, but it only takes into account victories....

Is this right? Why not use the logs in losses to include all opponents?

a) What's the point?
b) It's frigging difficult for Terminator games and I don't have that much time at the moment.

The point is most games end in losses. I see what you mean about terminator games though. I reckon a workaround would go something like this: Say we have a 6 player game. Then at the end, the log will have recorded 5 kills, and given out A, B, C, D and E points. If you won the game, then you recorded at least one kill, so lets say you got E points for it, from player 2. So you've got your ratio against player 2, just by taking E/20 (and times 1000 to make it the readable format you are using). Now if player 2 killed someone (player 3 for D points), then you can combine the two and find your ratio with player 3, by taking E*D/400. If player 2 didn't kill anyone, then you killed more than one person, so we can ignore player 2 and treat it as a 5 player terminator game.

Take some time off from this project if it's getting too big. We're all here to play Risk anyways, some people are just excited to use the full power of past game logs to make JR24 shut up. Or recruit some other people to help add functions. There's a usergroup memberlist.php?mode=group&g=192856 who enjoy all things computer related.

Elementary Maths is not the issue here, it's the code.
In order to work everything out (this includes chains of relative scores), the code has to be reworked a lot - that's the difficult part.
People were asking for this version today so I gave it to them.

I haven't clicked on your link yet - is it an invite to a new Usergroup?

Thank you very much!

Hi chipv,

I'm not really understanding how this relative kill rank is calculated.

I did a Map Rank pass for my Quad games, and I have only played 1 Arms Race! Quad: Game 2510984

We won 18pts for that game, which was brig, col, maj, maj vs col, maj, maj, maj

YET, Map rank for that map was calculated as nOOb Farmer (300)....?

Which means that for that game Map Rank thinks I had a score in excess of three times that of my opponents?

Pls explain.

Thanks, Ben

BENJIKAT IS DEAD wrote:Hi chipv,

I'm not really understanding how this relative kill rank is calculated.

I did a Map Rank pass for my Quad games, and I have only played 1 Arms Race! Quad: Game 2510984

We won 18pts for that game, which was brig, col, maj, maj vs col, maj, maj, maj

YET, Map rank for that map was calculated as nOOb Farmer (300)....?

Which means that for that game Map Rank thinks I had a score in excess of three times that of my opponents?

Pls explain.

Thanks, Ben

I missed that (lucky I'm redoing the calcs for losses).

Your point yesterday is the reason for this error.
Your teammates have not been counted and there is also an inexplicable subtraction which
does this: 18/20 = 0.9. Divide by 8/4 -1 = 0.9/3 = 0.3 x1000 = 300 = n00b Farmer.

Corrected calc:

72/20 = 3.6 divide by 4 = 0.9 0.9 x 1000 = 900

Fixed for next version, thanks for pointing that out, sorry.

So I'm guessing that Blitz was at (or very very near) the top of your Noob Farmer list.... due to the miscalculation of team scores. (He has over 3700 team games, which are a large proportion of his total).

Look forward to the fixed code, and the updated list which I think will more closely resemble the suggested culprits.

BENJIKAT IS DEAD wrote:So I'm guessing that Blitz was at (or very very near) the top of your Noob Farmer list.... due to the miscalculation of team scores. (He has over 3700 team games, which are a large proportion of his total).

Look forward to the fixed code, and the updated list which I think will more closely resemble the suggested culprits.

Yeah it will. People playing a lot of team games grossly put out, sorry again, people, should have refrained from making this available.
Also I think it does make sense to include the losses as well. The new stat will be relative opponent not relative kill.

btw I also think that your labels, noob Framer etc need to have their bands tweaked a bit.

I don't think 500 is low enough a figure to class as Noob Farming. For someone with a score of 3000, that threshold represents an average of 1500 pt opponents. Personally I would like to see it dropped to 400 or 350, but I guess we need to see the true calculations for MOBA / torcav / KLOBBER etc.


As an aside, do you know how to calculate which games actually have a bearing on the current score? i.e. After a certain number of games, given the same results against the same opponents, you would end up with the same score regardless (within reason) of the score you started with. A player's current score is therefore only the result of their last X games, with X varying greatly with a number of factors: game type, number of players and level of opposition being the most noteable.... Not sure where I'm going with this, but maybe it will strike a chord with you.

BENJIKAT IS DEAD wrote:. After a certain number of games, given the same results against the same opponents, you would end up with the same score regardless (within reason) of the score you started with.

Ah, but it would depend on the scores of everyone else in your group. Suppose you play the same 5 opponents over a large number of games, enough to move everyone into their expected score zone. If everyone started with 1000 points before you begin playing, then your score (and everyone else's) would be lower than if everyone was at 2000 before you began playing each other. You score (after a certain number of games) will be a function of ability and the average score of the people you play against. If you play everyone (or a wide variety of opponents), then the average should be near 1000, and your score should reflect ability. But if you only play certain ranks (high or low, since low ranks lose out due to the discrete structure of games and rounding), your score will be inflated.

As for the labels, they should be taken with a grain of salt. Anyone over a certain score will have difficulty playing games against people with 3/4 of their score, and so will move down a lot. It should be clear that players with a high score will have a lower rating, and they really should if they're playing a variety of opponents.

I'd move levels so that they're centred around 1000. That is, call 800-1200 the average, then 0-400 is worst, 400-800 is low, 1200-1600 is high, and 1600+ is best. I haven't seen anyone clear 1300, nevermind 1600, and 80% of your score is reachable for most players, while anyone lower than 40% is pretty obviously just picking weak opponents. For a 3000 point player, lower than 40% would mean for every Major+ they play, they also played a cook with 400-. There are only 53 active cooks in that range, and most only get 4 games at a time!

Totally agree on levels, BA.

Almost done with the losses calculations, and what a absolute nightmare. Deadbeats didn't always get scored which makes both
team and terminator calcs a nightmare. Just running through some final tests on terminator losses recursive calcs and should be done. Thanks for all the help, by the way, in fact it was you who first posted this idea a while ago.

chipv wrote:Thanks for all the help, by the way, in fact it was you who first posted this idea a while ago.

Thanks for implementing it! I was using Map Rank for my own primitive version to run on some suspect individuals, but I had to go through each team size and single player size individually, and even then it only counted the opponent who won. This makes things a lot easier, plus everyone in the community can check values, which they couldn't do (easily) with my method.

BaldAdonis wrote:

BENJIKAT IS DEAD wrote:. After a certain number of games, given the same results against the same opponents, you would end up with the same score regardless (within reason) of the score you started with.

Ah, but it would depend on the scores of everyone else in your group. Suppose you play the same 5 opponents over a large number of games, enough to move everyone into their expected score zone. If everyone started with 1000 points before you begin playing, then your score (and everyone else's) would be lower than if everyone was at 2000 before you began playing each other. You score (after a certain number of games) will be a function of ability and the average score of the people you play against. If you play everyone (or a wide variety of opponents), then the average should be near 1000, and your score should reflect ability. But if you only play certain ranks (high or low, since low ranks lose out due to the discrete structure of games and rounding), your score will be inflated.

You missed my point slightly - take my entire game history - how many of the last games "actually" count against my current score - i.e. with ALL other factors remaining the same, how many games back into my history do you need to go in order to end up at the same current score regardless of what score I had at that point.

I played Magic the Gathering competitively for some years, and a similar relativistic scoring system was used for that... one of their statisticians figured out that after 200 or so matches the previous record was lost within the rounding.

So, what I'm saying is that our current score is solely a function of our success or otherwise in the last X games - just wondering whether anyone is clever (and time-rich) enough to figure out the value of X?

Ok, phew, just finished - anything else to go in quickly?

chipv wrote:Ok, phew, just finished - anything else to go in quickly?

Get a handful fo people to test it first this time ?

BENJIKAT IS DEAD wrote:

BaldAdonis wrote:

BENJIKAT IS DEAD wrote:. After a certain number of games, given the same results against the same opponents, you would end up with the same score regardless (within reason) of the score you started with.

Ah, but it would depend on the scores of everyone else in your group. Suppose you play the same 5 opponents over a large number of games, enough to move everyone into their expected score zone. If everyone started with 1000 points before you begin playing, then your score (and everyone else's) would be lower than if everyone was at 2000 before you began playing each other. You score (after a certain number of games) will be a function of ability and the average score of the people you play against. If you play everyone (or a wide variety of opponents), then the average should be near 1000, and your score should reflect ability. But if you only play certain ranks (high or low, since low ranks lose out due to the discrete structure of games and rounding), your score will be inflated.

You missed my point slightly - take my entire game history - how many of the last games "actually" count against my current score - i.e. with ALL other factors remaining the same, how many games back into my history do you need to go in order to end up at the same current score regardless of what score I had at that point.

I played Magic the Gathering competitively for some years, and a similar relativistic scoring system was used for that... one of their statisticians figured out that after 200 or so matches the previous record was lost within the rounding.

So, what I'm saying is that our current score is solely a function of our success or otherwise in the last X games - just wondering whether anyone is clever (and time-rich) enough to figure out the value of X?

I understand what you're saying, but would it be a set number of games X for everyone? Somehow I think it would be relative to each player... but I could be wrong.

FabledIntegral wrote:I understand what you're saying, but would it be a set number of games X for everyone? Somehow I think it would be relative to each player... but I could be wrong.

Yes X would vary for each player, and would also vary over time, due to the wide array of game choices we have.

The 6-8 player game guys would have a far far lower number than the 1v1ers.

BENJIKAT IS DEAD wrote:

chipv wrote:Ok, phew, just finished - anything else to go in quickly?

Get a handful fo people to test it first this time ?

Ok, done, beta uploaded, although how much response I will get remains to be seen.

chip, I was wondering if you could change the value to a decimal with a fixed three decimal places. This would not be any different than what you have now, other than the three orders of magnitude. The reason I ask this, is because the first time I looked through the numbers, I thought these were scores. If that does not work, can you put XXXX/1000, just so that it is clear that this is not a score?

lancehoch wrote:chip, I was wondering if you could change the value to a decimal with a fixed three decimal places. This would not be any different than what you have now, other than the three orders of magnitude. The reason I ask this, is because the first time I looked through the numbers, I thought these were scores. If that does not work, can you put XXXX/1000, just so that it is clear that this is not a score?

Not a problem - my preference also, wanted to emphasize that a small fraction is quite a large difference but people get that, so, yes.

BENJIKAT IS DEAD wrote:I played Magic the Gathering competitively for some years, and a similar relativistic scoring system was used for that... one of their statisticians figured out that after 200 or so matches the previous record was lost within the rounding.

So, what I'm saying is that our current score is solely a function of our success or otherwise in the last X games - just wondering whether anyone is clever (and time-rich) enough to figure out the value of X?

I see what you mean now, but I reckon it'll be too heavily dependent on game style, so unless your player has some exclusive specialty, you won't be able to figure it out.

Suppose you started over (that never happened, did it? ) and played 2 player games against people with exactly your score, and compare to player X, who exclusively plays 1v1s against someone with half their score. After 50 games, you decide to try his idea, and play people with half your score. It would take you 950 games to get your ratio down to his level (within 5%, which is where I would put rounding errors, due to integer point values).

I figure people who have more opponents would have lower values, so 1000 is probably a good ceiling estimate.

BaldAdonis wrote:

BENJIKAT IS DEAD wrote:I played Magic the Gathering competitively for some years, and a similar relativistic scoring system was used for that... one of their statisticians figured out that after 200 or so matches the previous record was lost within the rounding.

So, what I'm saying is that our current score is solely a function of our success or otherwise in the last X games - just wondering whether anyone is clever (and time-rich) enough to figure out the value of X?

I see what you mean now, but I reckon it'll be too heavily dependent on game style, so unless your player has some exclusive specialty, you won't be able to figure it out.

Suppose you started over (that never happened, did it? ) and played 2 player games against people with exactly your score, and compare to player X, who exclusively plays 1v1s against someone with half their score. After 50 games, you decide to try his idea, and play people with half your score. It would take you 950 games to get your ratio down to his level (within 5%, which is where I would put rounding errors, due to integer point values).

I figure people who have more opponents would have lower values, so 1000 is probably a good ceiling estimate.

You've realised the reason I'm asking on a personal level now!!

Is there a way I can do searches without looking at a particular person? I want to check out which maps favour the first player the most in 2 player games.

BaldAdonis wrote:Is there a way I can do searches without looking at a particular person? I want to check out which maps favour the first player the most in 2 player games.

No (it is an easy change though, could let you know) but I took that decision early to limit the information requested from the server. You would have to search every single 2 player game to get your info. There is a playing first stat in map rank already but as you say, player specific.

chipv wrote:No (it is an easy change though, could let you know) but I took that decision early to limit the information requested from the server. You would have to search every single 2 player game to get your info. There is a playing first stat in map rank already but as you say, player specific.

I was thinking map specific, but even still, the ones I wanted were New World (5000 games), Age of Merchants (7000) and Waterloo (3000). I bet this app is already making a noticeable bump in the database requests.

BaldAdonis wrote:

chipv wrote:No (it is an easy change though, could let you know) but I took that decision early to limit the information requested from the server. You would have to search every single 2 player game to get your info. There is a playing first stat in map rank already but as you say, player specific.

I was thinking map specific, but even still, the ones I wanted were New World (5000 games), Age of Merchants (7000) and Waterloo (3000). I bet this app is already making a noticeable bump in the database requests.

No I have over 7600 games played and I can do my full Map Rank.

ParadiceCity9 wrote:

BaldAdonis wrote:

chipv wrote:No (it is an easy change though, could let you know) but I took that decision early to limit the information requested from the server. You would have to search every single 2 player game to get your info. There is a playing first stat in map rank already but as you say, player specific.

I was thinking map specific, but even still, the ones I wanted were New World (5000 games), Age of Merchants (7000) and Waterloo (3000). I bet this app is already making a noticeable bump in the database requests.

No I have over 7600 games played and I can do my full Map Rank.

I'm not saying you can't do it, just that if it were an option, and anyone could check any map (or every map), then there would be a big spike.