highest/lowest relative rank

[roadwarrior]

Equalitarian (0.902)
I was curious and found my equalitarian rating =0.902 today.

Can someone explain how the rating dropped from 0.932 only from 2 or 3 days ago. I only just completed 2 more games since where the ranks of my opponents were averagely between 0.85 to 1 relative to mine. Can 2 “similarly ranked games” make such a big difference or were there other factors such as a readjustment in the way the calculations work? Note: I ran version 1.5.5, so maybe that was the reason why.

Anyway, my current CC score is 3302 and if my understanding is correct, that should work out to 2978 relative rank.

To be honest, these past 2 or 3 months. I paid scant attention to the CC scoring system but with this new development, I am curious to know how many are actually above 3000 on a readjusted basis.

[qeee1]

version 1.5.5 affected a bug which meant that terminator games weren't being counted correctly. You've played a good few terminator games (68) so I expect this is what caused the difference and not the two extra games you played. chipv is the one who'd know for sure though, he coded the script.

Blitz, I'll get running through the names when I get back to my home computer.

[chipv]

Equalitarian (0.902)
I was curious and found my equalitarian rating =0.902 today.

Can someone explain how the rating dropped from 0.932 only from 2 or 3 days ago. I only just completed 2 more games since where the ranks of my opponents were averagely between 0.85 to 1 relative to mine. Can 2 “similarly ranked games” make such a big difference or were there other factors such as a readjustment in the way the calculations work? Note: I ran version 1.5.5, so maybe that was the reason why.

Anyway, my current CC score is 3302 and if my understanding is correct, that should work out to 2978 relative rank.

To be honest, these past 2 or 3 months. I paid scant attention to the CC scoring system but with this new development, I am curious to know how many are actually above 3000 on a readjusted basis.

Before 1.5.5 there were some terminator logs (with deadbeats and objectives) that caused a miscount because of their alternate format so 1.5.5 is actually now accurate. So don't worry about pre-1.5.5 results.

[roadwarrior]

This is a nice idea.

If I understand your methods, your calculation is affected by all games that we ever played and the weights of each game is the same. I like to seek more information regards the methods of calculation because it did not seem to be explicitly mentioned except for your brief note:
“Note that the score is calculated based on your score at each game end time, not your current score”

Take a hypothetical case:
For example, Game#1 is weighted the same (from what I can tell or not) as the most recent game played. When a new player starts with CC, he naturally would play with players around his rank and progresses to higher ranks over the course of time.

Say that a high ranker 3000 just finished a game and the average score of all opponents was 3000, so this equals to relative rank of 1 based on his current end game CC score of 3000. I do not know if his current end game score is benchmarked against the first game (assuming the average score of all his opponents then were 1000), will the equalitarian rank(ER) for game #1 be re-calculated as 1000/3000= 0.33333 because of the most recent end game score?

If both games are equally weighted, then the average ER is around 0.66666. This of course is not accurate since our hypothetical player plays with similar ranked players.

So I wish to clarify that the above is not the case and that aside, you may also consider to attach more weights to most recent games and lesser weights to long past games. It is the current profile of a player that should matter more, and much less a year ago. After all, players do change and may altogether be a different player compared to a year ago.

Perhaps the most recent 40% games should be given 70% weightage and the mid 40% =25% weightage and the first 20% games only 5% weightage.

By the way how do you calculate team game scores?

[chipv]

This is a nice idea.

If I understand your methods, your calculation is affected by all games that we ever played and the weights of each game is the same. I like to seek more information regards the methods of calculation because it did not seem to be explicitly mentioned except for your brief note:
“Note that the score is calculated based on your score at each game end time, not your current score”

Take a hypothetical case:
For example, Game#1 is weighted the same (from what I can tell or not) as the most recent game played. When a new player starts with CC, he naturally would play with players around his rank and progresses to higher ranks over the course of time.

Say that a high ranker 3000 just finished a game and the average score of all opponents was 3000, so this equals to relative rank of 1 based on his current end game CC score of 3000. I do not know if his current end game score is benchmarked against the first game (assuming the average score of all his opponents then were 1000), will the equalitarian rank(ER) for game #1 be re-calculated as 1000/3000= 0.33333 because of the most recent end game score?

If both games are equally weighted, then the average ER is around 0.66666. This of course is not accurate since our hypothetical player plays with similar ranked players.

So I wish to clarify that the above is not the case and that aside, you may also consider to attach more weights to most recent games and lesser weights to long past games. It is the current profile of a player that should matter more, and much less a year ago. After all, players do change and may altogether be a different player compared to a year ago.

Perhaps the most recent 40% games should be given 70% weightage and the mid 40% =25% weightage and the first 20% games only 5% weightage.

By the way how do you calculate team game scores?

Each calculation is done at the exact time the points are added/deducted. The calculation is done by reading the game logs, which is the only place where the accurate relative ranks can be deduced at the time of points reassignment.
For each gain/loss = loser/winner * 20 we can derive relative rank = gain/loss / 20.

So that means there is no weighting at the moment.

Game #1 average score of opponents = 1000 => RR = 1
Game #200 average score of opponents = 3000 when your score is also 3000 => RR = 1

Team scores are done just like the points gain/loss are - calculation is done by teams so

RR = Team relative rank / number of players per team

[roadwarrior]

Thank you for your quick reply.

I am curious if my proposed weightage will impact the current calculations signficantly. There are many examples from time series data analysis that utilise such weightage tools to improve the reliability/accuracy of results. We are dealing with datas and observations across time (people and playing preferences of course do change). I would suggest you should pay more attention to that aspect in order to improve the reliability of your tools.

Cheers.

[chipv]

Thank you for your quick reply.

I am curious if my proposed weightage will impact the current calculations signficantly. There are many examples from time series data analysis that utilise such weightage tools to improve the reliability/accuracy of results. We are dealing with datas and observations across time (people and playing preferences of course do change). I would suggest you should pay more attention to that aspect in order to improve the reliability of your tools.

Cheers.

The current calculation is quantitative. Weighting is qualitative and highly subjective.
I'm not sure where you see unreliability in the current implementation, although perhaps you are looking for another word.

I've had a look at your proposal. What are those numbers based on?
At the moment the calculation doesn't make any assumptions about relevance of data across time, it assumes every game is equally relevant but...

There has been a request for something similar that only measures the last predetermined proportion of games, perhaps that is a middle ground.

[djt5483]

someone check me mine is prolly pretty low

[roadwarrior]

"There has been a request for something similar that only measures the last predetermined proportion of games, perhaps that is a middle ground"

This idea basically just truncates the data set giving the section not selected 0% and the selected set, equal weightage to each observation. We are back to square 1.

IF you wish to truncate out the first parts, then this is saying that the after a certain period of time, those games do not give any useful information so that they can be taken out of the dataset. However, you are still faced with deciding how to apportion weights to observations still left in your dataset.

Like I explained, people change overtime and for most, the change is a process.....it is not an abrupt thing that is to be modelled by a truncation like the above. For example, by giving equal to weightage to each observation, this is implying that no one observation is anymore informative than the other or equally useless. In this case, then the usefulness of your model becomes severely limited. It is just calculating for calculating sake and anyone can do that.

If you assume that the most recent games gives more relevant information about who the actual player are, then you want to give much higher weightage to the most recent ones but less to the less recent ones. This is subjective and my numbers are just examples and another set of numbers can just as well be used to illustrate the point. The point is to have an informed judgment about the weights and that is what that distinguishes a good model that can be relied upon. So what other words would you suggest using?

[chipv]

"There has been a request for something similar that only measures the last predetermined proportion of games, perhaps that is a middle ground"

This idea basically just truncates the data set giving the section not selected 0% and the selected set, equal weightage to each observation. We are back to square 1.

IF you wish to truncate out the first parts, then this is saying that the after a certain period of time, those games do not give any useful information so that they can be taken out of the dataset. However, you are still faced with deciding how to apportion weights to observations still left in your dataset.

Like I explained, people change overtime and for most, the change is a process.....it is not an abrupt thing that is to be modelled by a truncation like the above. For example, by giving equal to weightage to each observation, this is implying that no one observation is anymore informative than the other or equally useless. In this case, then the usefulness of your model becomes severely limited. It is just calculating for calculating sake and anyone can do that.

If you assume that the most recent games gives more relevant information about who the actual player are, then you want to give much higher weightage to the most recent ones but less to the less recent ones. This is subjective and my numbers are just examples and another set of numbers can just as well be used to illustrate the point. The point is to have an informed judgment about the weights and that is what that distinguishes a good model that can be relied upon. So what other words would you suggest using?

Agree with comments about truncation, this was only a suggestion from someone.

Ok, one problem with weighting relative rank is that by giving more weight to recent games, this further accentuates the inherent bias against the higher ranked players (and lowest as well).

The higher ranked players certainly over more recent games will have less 'choice' about the relative rank they play and so their relative rank is already going to be lower than a cook for example simply because there are less players at or above their rank.
Giving the recent games more weight will accentuate this.

The 'model' shows what relative rank a player may choose to play. For example someone who picks on lower ranks is quite obvious from this present calculation. It is this pattern which may well be constant over time. Weighting would obscure this not clarify this.
I'm not sure what result you are looking for by giving the numbers an albeit well-informed weighting because the statistic in question is a relative one, not an absolute one.

[BENJIKAT IS DEAD]

roadwarrior is describing in a different way the same thing I was suggesting a few pages back.

Can't help thinking it's very tricky to figure out ... if every game were 1v1 it would be much easier.

[chipv]

roadwarrior is describing in a different way the same thing I was suggesting a few pages back.

Can't help thinking it's very tricky to figure out ... if every game were 1v1 it would be much easier.

It's impossible at present.

The games can be ordered by time started but not time finished.
I cannot provide weighting unless this is made possible.

[Me-Da-MiNoRiTY]

YAY another forum i didn't know i was mentioned on ^^

Hehe these are interesting statistics

[chipv]

Hold on, this might be possible - the game logs are timestamped.

[Blitzaholic]

Blitz, I'll get running through the names when I get back to my home computer.

ok, thx qeee1

[Blitzaholic]

someone should run

georgerx7di
khazalid
roadwarrior
jasperjack
slamhog
andrewb
enhenrickson
genghis khan ca
gibbom
dangermouse
collett6
comic boy
billy7
robinette
chefo
clapper011
mitriclark
pedronicus
louisp141
loudawg
sully800
golfer1024
big yuma ripper
j33j
rl_orange
yunnick
dupa
marriedman
wertz407
osujacket
virual
david_wain
nephilim
thanatose
xtratabasco
fireside poet
wacicha
eye84free
ptlowe
alstergren
jaydog
djt5483


gibbom may be # 1




update all those and I got like 30 more for you, but dont want you to get too overwhelmed



no doubt many of these names will make one of the top 25 lists

[rebelman]

new top 10 entry - rebelman 1.257 i love playing guys higher ranked than me

and the lower my score goes the higher this figure will get

[Blitzaholic]

new top 10 entry - rebelman 1.257 i love playing guys higher ranked than me

and the lower my score goes the higher this figure will get

many do, but when your close the top for long extended periods of time and you have an average score of 4000 or over a year str8, when the cc average is like 1000, it cripples you, so this system is a bit flawed for the higher ranks, also, battle royal winners tend to really mislead the numbers as well.


yo qeee1, have you not been home in the last 24 hours? any updates on those names?

[Fruitcake]

new top 10 entry - rebelman 1.257 i love playing guys higher ranked than me

and the lower my score goes the higher this figure will get

many do, but when your close the top for long extended periods of time and you have an average score of 4000 or over a year str8, when the cc average is like 1000, it cripples you, so this system is a bit flawed for the higher ranks, also, battle royal winners tend to really mislead the numbers as well.

I agree Blitz, but you are missing the point.

Rebeleman may have a high relative ranking number, but this still makes his relative score just 1104, or the relative rank of Corporal. Your score of 3710 and relative score of 0.560, gives you a relative rank of Major, QED the differential although narrowed is still large. If you check back to my list from a few days back, it was obvious that the top players were, in fact, exactly that, top players. Anyone with a low score can grab a reasonably high relative rating by taking on Senior ranks, the key is, and always will be, to keep the relative score as high as possible during ones rise through the scoreboard.

Regarding battle royal winners....well there will always be the exception that proves the rule.

What would be interesting, would be to have 2 scoreboards, one based on simple points gained, the other with the relative rating factored in. I am sure this would stop a lot of the moaning about top players, as, in my opinion, the leader page would be very much the same. It would also be quite a grand prize for a Conquerer to hold both pages at once!

[qeee1]

Guys, I'm not gonna maintain this anymore. I'm just a fun loving guy; I was in it to see a couple of the top ranks get egg on their faces, for the controversy, for the chicks. Now it's become tedious and I'm bailing, because that's the kinda person I am.

Someone else can start a thread if they want.

[BENJIKAT IS DEAD]

Guys, I'm not gonna maintain this anymore. I'm just a fun loving guy; I was in it to see a couple of the top ranks get egg on their faces, for the controversy, for the chicks. Now it's become tedious and I'm bailing, because that's the kinda person I am.

All it showed us was that torcav and MOBA pick on noobs and that Blitz once had a score that was obscenely higher than everyone else's - and we already knew that!

From my perspective this list was encouraging me to play less public games - something that is so often complained about - guess we can't win!

[roadwarrior]

chipv wrote:
"Ok, one problem with weighting relative rank is that by giving more weight to recent games, this further accentuates the inherent bias against the higher ranked players (and lowest as well).

The higher ranked players certainly over more recent games will have less 'choice' about the relative rank they play and so their relative rank is already going to be lower than a cook for example simply because there are less players at or above their rank.
Giving the recent games more weight will accentuate this.

The 'model' shows what relative rank a player may choose to play. For example someone who picks on lower ranks is quite obvious from this present calculation. It is this pattern which may well be constant over time. Weighting would obscure this not clarify this.
I'm not sure what result you are looking for by giving the numbers an albeit well-informed weighting because the statistic in question is a relative one, not an absolute one."

I think your tool works in picking out noob farmers very nicely. However, your tool is capable of greater potential. To be honest, what I am interested to know is a rating that can tell me who are the good players. The problem with the current CC system is that I cannot consider it to be a reliable system for a good first indication of a player’s abilities.

I am hoping to share a framework with you that can extend the scope of your tool to indicate a truer profile of a player and hopefully can give more useful information about his abilities. I notice that there is currently a lot of “background noise” that ought to be removed from the dataset to arrive at a player’s current profile accurately.

Take for example, Torcav2. He is an example of a type of player that has changed his play style overtime. He was a very different player when he first started but all he does now is to farm noobs in 1vs 1 freestyle games.

Therefore his games from before do not indicate the type of the player he is today and should be excluded from the dataset. If you exclude these early games, his RR should even be lower which is exactly the true ranking of such a player.

According to the CC menu, here are all possible game variants available to choose from:
#of players =7 options
Game type= 6 options but if you consider doubles, triples, quads as a single team option, then 4 options= Standard/Terminator/Assassin/Team
Play order =2 options= Freestyle and Sequential
Bonus cards= 3 options
Fortifications= 3 options
Fog= 2 options
Round length= 2 options.

So there are at least 2016 variants to choose from (that many!). Here is a suggested framework how to determine the weights for games based on current playing preferences of the player himself. You can expect players tend to favor few variants eventhough there are so many to choose from.

Using the most recent 40% of games finished and then apportion the weights according to the frequency of the preferred variants of the player himself and taking the case of Torcav2 again.Total=1184 games so 40% equals the most recent 480 games (approx the most recent 5 pages of his games history). In the case of Torcav2, he only favours 1 variant which is 1vs1 freestyle against noobs and that simplifies the illustration! The weight of 1vs1 freestyle should be 100% in Torcav’s example. This means, no other variants that he had ever played should be entered into the calculations of his current RR since they do not indicate the type of the player he is right now and therefore irrelevant and given weights of 0.

However, this does not mean to truncate the first 60% of the dataset since you still want to look for 1vs1 freestyle that he may have played in the first 60% of his games to calculate. If they exist in the first 60% of his games, they are still relevant and should be scanned for his current RR calculations.


This will give a truer impression of a player’s actual profile and in the case of Torcav’s, his actual RR should even be lower than your current calculations with my suggested framework.

My suggestion to apportion game variant weights based on play preferences for the most recent 40% games is only an example. You may consider a range from 30% to 50% or whatever it is. You just have to rely on rule of thumb to see which results are more convincing and reasonable given what you know based on the examples of a few players.

One concern you raised is time bias that affects high rank negatively if we apportion higher weights to more recent games. However, I think that the issue is dealt with my suggested framework since the most recent games only serves to lay out the basis for weights to be apportioned to all games in the dataset. This means a game will not lose its relevancy no matter when it was played so long as the player continues to play the variant currently.

I hope that you will continue this good work

[lancehoch]

I get the feeling this would cause a lot of strain on the server. I am not positive on this, but it just seems that you would be making a lot of server requests:

• player profile (for number of games played)
• game finder page (to look up games)
• data request on each of the most recent x% of games played (for what options were used)
• log request on all games (to actually get relative ranks from that game)

Just a thought.

[Blitzaholic]

Guys, I'm not gonna maintain this anymore. I'm just a fun loving guy; I was in it to see a couple of the top ranks get egg on their faces, for the controversy, for the chicks. Now it's become tedious and I'm bailing, because that's the kinda person I am.

All it showed us was that torcav and MOBA pick on noobs and that Blitz once had a score that was obscenely higher than everyone else's - and we already knew that!

From my perspective this list was encouraging me to play less public games - something that is so often complained about - guess we can't win!

exactly and qeee1 it proves nothing, plus, i had many more names than those lists that would make many on that orignal list not even be on there, you only did a few dozen names, and said you would do others and never did once you saw the numbers, lied you did, plus I had 50 more names to show you that would alter your list.

Ask around with all of the most respected and feared on the CC site, most will tell you if they honest who is who and who is super in conquer club.

[qeee1]

Guys, I'm not gonna maintain this anymore. I'm just a fun loving guy; I was in it to see a couple of the top ranks get egg on their faces, for the controversy, for the chicks. Now it's become tedious and I'm bailing, because that's the kinda person I am.

All it showed us was that torcav and MOBA pick on noobs and that Blitz once had a score that was obscenely higher than everyone else's - and we already knew that!

From my perspective this list was encouraging me to play less public games - something that is so often complained about - guess we can't win!

exactly and qeee1 it proves nothing, plus, i had many more names than those lists that would make many on that orignal list not even be on there, you only did a few dozen names, and said you would do others and never did once you saw the numbers, lied you did, plus I had 50 more names to show you that would alter your list.

I played against the very best on the cc site in singles which are sjnap, comic boy, scott-land, joecoolfrog, benjikatisdead, maniacmath17, poo-maker, thota, and so many others and I do well and hold my own easy although I do not play them much, ask anyone of them, I also played with and against the best team players on the site evidenced by clan challenges and tournies and do very well against all of them and hold my own really well, and also i open my game up to ALL players in public games with none on my ignore list except a couple who are no good anyways and challenged just about all and played just about anyone and everyone and at one time or another and had success off most of them including the game that you and rev challenged and called me and khazalid out and we smashed you and you made some horrific plays which shows me you are not to good in team games, so your little retarded list is a joke. Plus I play all maps and all styles, a variety of everything, not just one easy way and thats it, its too simple to be only good at one style and thats all you play is singles most of the time, where I am well rounded. As far as teams goes it does help that I got some of the very best on the cc site all in the clan I am in, so they all deserve a lot of the credit. Plus you been out and gone for long long time and much has changed, field of competition much stronger, and there has been several polls over the last year voting the cc community who is best, and my name was up there quite often and even won some of those polls. Now, i am not claiming to be the best, like scott-land and others said, it is just way too hard to decide that with all the versatility. I am claiming i am one of the very best or top dozen on the cc site world for well rounded all around versatile play and I dont think anyone who has played with me or against me would argue with that. So, your little egg on your face list is garbage, so do not try to make an attempt to smear my name with your bias and envy and proves NOTHING and makes you look gross. ALL that play with me or against me on the battlefield know this, so I need to prove nothing to nobody especially you. Yes I am a little pissed off cause of your comments and an attempt to assault my name, but I am in full agreement with benjikat is dead, it proves nadda, zip, zilch. so, since you have left the site and been gone for a year or more, you think by putting this little stat together shows you something, again, it proves nothing, ask around with all of the most respected and feared on the CC site, most will tell you if they honest who is who and who is super in conquer club, experience is the best teacher. dont expect to make an attack on me without me coming back to make you look like an absolute joke.

But Blitz I didn't attack you