May GhostRatings Published

Forum rules
This forum is limited to topics relating to the game Diplomacy only. Other posts or topics will be relocated to the correct forum category or deleted. Please be respectful and follow our normal site rules at http://www.webdiplomacy.net/rules.php.

Post a reply

Confirmation code
Enter the code exactly as it appears. All letters are case insensitive.
Smilies
:points: :-D :eyeroll: :neutral: :nmr: :razz: :raging: :-) ;) :( :sick: :o :? 8-) :x :shock: :lol: :cry: :evil: :?: :smirk: :!:
View more smilies

BBCode is ON
[img] is ON
[flash] is OFF
[url] is OFF
Smilies are ON

Topic review
   

If you wish to attach one or more files enter the details below.

Expand view Topic review: May GhostRatings Published

Re: May GhostRatings Published

by damo666 » Sun Jul 29, 2018 10:49 am

Thanks

Re: May GhostRatings Published

by ghug » Sun Jul 29, 2018 12:53 am

I'll get it all up to date at once once August starts. Classic is going to require more time to fix, and I sadly don't have much of that right now.

https://github.com/negaffney/webDiplomacyRatings is public if anyone would like to take a look.

Apologies to all, it's been a busy summer.

Re: May GhostRatings Published

by damo666 » Sat Jul 28, 2018 5:47 pm

What is the story with June ratings and how is it going clearing non-classic variants from non-live classic?

Re: May GhostRatings Published

by jengamaster » Tue May 22, 2018 9:38 pm

Top 25 in the gunboat category. Well. I am ready to retire now. Where is my gold watch?

Re: May GhostRatings Published

by ghug » Tue May 22, 2018 6:10 pm

My intent right now is to post the one spreadsheet with columns for (new) overall, AvF, and GvI ratings, sorted by the first. I'm still open to tweaks on that if anyone has thoughts.

Re: May GhostRatings Published

by MadMarx » Tue May 22, 2018 5:44 pm

ghug, will you clarify, is it your intent to post AvF, GvI *and* a weighted / combined 1v1 GR each and every month, ‘til death do you part?!

Re: May GhostRatings Published

by ghug » Tue May 22, 2018 1:54 pm

There actually may be an issue with country weighting in that the GR data dump doesn't report country ID. It might give them to us in order, but I'm unsure.

Re: May GhostRatings Published

by Yonni » Tue May 22, 2018 11:06 am

CCR wrote:
Tue May 22, 2018 10:09 am
@Yonni, again, it doesn't matter if GvI is separated from FvA, as long as you keeep an overall 1v1 rank. Wonder there are maybe a dozen 1v1 maps available on vdiplomacy.net and in the ideal world we could see your elo method applied elsewhere ;)
I know it's a massive undertaking and not likely at the top of the priority list but, if GR ever gets site integration, keeping a separate rating for each variant would definitely make sense. It' just real clunky with the old spreadsheet.

Re: May GhostRatings Published

by CCR » Tue May 22, 2018 10:09 am

@Yonni, take a closer look to the idea above - weighing powers! On GvI it is very predictable: G wins, the winner takes fewer points, the loser loses the same fewer points; I wins, the winners takes more points, the loser the same loses more points. Next step, apply it to all maps (I wonder an italian solo would count more than a french solo, etc etc).

@Yonni, again, it doesn't matter if GvI is separated from FvA, as long as you keeep an overall 1v1 rank. Wonder there are maybe a dozen 1v1 maps available on vdiplomacy.net and in the ideal world we could see your elo method applied elsewhere ;)

Re: May GhostRatings Published

by MadMarx » Tue May 22, 2018 7:19 am

ghug wrote:
Sat May 19, 2018 4:30 pm
I'm also selfishly not in favor of increasing the number of GR categories above where it is because there's only so much automation that can go into that process as long as ratings aren't integrated with the site. I do think there is a decent amount of logic in grouping the two together, as they do have obvious fundamental similarities, and I think both are worth capturing.

I'm happy to run the two separately as a one-off for the curious.
The spreadsheet you linked is just a “one-off for the curious.” You are “not in favor of increasing the number of GR categories above where it is.” The actual
GR link that gets published shows only one 1v1 rating, so the proposed change, as I understood it, was *just* a combined AvF / GvI revised rating with neither AvF or GvI individial ratings. If you’re intention has been to post all three 1v1 ratings then we’ve been wasting our time chatting about this ever since. COME ON, MAN!! :-)

Re: May GhostRatings Published

by ghug » Tue May 22, 2018 6:22 am

Did you look at the sheet I posted? There are columns right there for the individual scores.

Re: May GhostRatings Published

by MadMarx » Tue May 22, 2018 5:44 am

The hostility is all in good fun, exaggerating in an attempt to make a point, but mostly to keep discussion going, but of course it has its flaws, so I’ll let it go.

You do confuse me though, you said the new GR would replace the old, and that you only want one 1v1 category, and now guys say the AvF and GvI individual GRs will *not* be hidden, which is it?! I couldn’t care less about a combined GR if we get the individual GRs as well, but that isn’t my understanding, regardless of what’s in you “one off” (or however you worded it) spreadsheet...

I think GvI has value, just different value, and when you combine two things that are different you end up with something that is not very useful.

How did you miss that I want AvF first, GvI second, and a combined third, did you not notice GvI in second right there?!

Re: May GhostRatings Published

by ghug » Tue May 22, 2018 5:19 am

I hardly think it's fair to classify anything Mike has said is elitist. He believes in the value of both variants, and he's correctly pointed out that this solution addressees the initial issues you raised, as well as new ones you're raising like individual scores being "hidden" when they are in fact right there in the spreadsheet. You disagree that GvI has value, or that they correlate (there's about a 70% R^2 value between individuals' independent scores in the two variants, which I'd clarify as statistically significant, but you do also have a point that they aren't identical skills). That's no reason to get hostile.

I appreciate that you feel strongly about this, but you seem to be the only person who does feel the need to eliminate GvI from the rankings entirely.

Re: May GhostRatings Published

by MadMarx » Tue May 22, 2018 4:58 am

CptMike wrote:
Mon May 21, 2018 7:23 am

You can't say if promoting something that would serve only 0.1% rather than the 99.9% is good or not ?

MadMarx, if there are some versions of the game that you don't want to play, why don't you just avoid playing them ? Feel free.

Ghug's proposal solves all the points that MadMarx put forward initialy : the new global GR based on the GR's of GvI and FvA, calculated separately. Being better at one or the other variant doesn't influence the result for the other variant any more.

Global GR is not just the sum of both divided by two. It weighs both individual GR by the number of games played by each player for each variant, meaning that if a player doesn't play many GvI or play them less often will not harm his global score at all... (eg FvA: 500 with 45 games and GvI 100 with 5 games would give a global GR of 460)
Personally, I would serve the 99.9%, but I’m trying to keep an open mind even though some of the 0.1% seem extremely adamant about championing themselves.

I will play whichever games I choose, regardless of GR, I play for fun not for GR. That said, your attitude is extremely elitist, encouraging others to not even try if they don’t go about it exactly as you deem to be the one and only way to do it. Furthermore, I have every right to champion the best possible GR system for everyone, GR is meant to engage more users and pull people into games/webDip. You are free to fight for exclusivity, I will continue to fight for inclusivity. I have provided multiple solid arguments, backed up by multiple other informed users, and your response is “if you don’t like how it is, then leave/quit”? Change takes a bit of tenacity, and even if change doesn’t take place, it’s still worth fighting for if you think something is broken and/or can be improved significantly. You have yet to provide a single worthwhile argument in my opinion, but I will keep reading what you and everyone else has to say.

A global GR doesn’t solve all the problems I put forward initially, since the individual AvF and GvI GRs would be hidden, at least one individual GR would need to be posted to solve all the problems.

How is a global GR of 460 *not* harmful to an actual AvF GR of 500, it is significantly less (40 points is the difference between top 50 and top 75, nearly the difference between top 100 and top 200) ? How is a global GR of 460 *not* harmful to an actual GvI of 100? I was seeded as one of the very best GvI players for the GvI tournament without ever having played a GvI game, how does that not give me an unfair advantage, and screw over dozens of others that had earned the right to a better seeding in that tournament?!

Finally, if anyone felt my attitude was a little intense during any of that, I certainly considered toning it down, but I’m intentionally trying to encourage a bit of discussion, please forgive me if I’ve stepped on your toes, though most importantly, voice your opinion!

Re: May GhostRatings Published

by Yonni » Mon May 21, 2018 1:53 pm

My 2 cents:

I think the purpose of overall GR is give us a fun all-inclusive rankings system to compete in and GR sub-categories are to suit the more niche needs of GR (i.e. tournament seeding, comparing specific skills of players). With that in mind, I think there is more of a want and application for FvA and GvI to be separated.

p.s. It's Elo, not ELO. It's not an acronym.

Re: May GhostRatings Published

by CptMike » Mon May 21, 2018 11:10 am

Mercy wrote:
Mon May 21, 2018 10:42 am
I can imagine that that last part would be a lot of work to program, so we can be less precise and just try to guess ourselves what standard deviations should be assumed.
That's another topic than the global GR but that's an interesting one. I think there is an easy way to respond to the "unbalance" issue for what concerns ELO points.

Let's assume that we currently have on the whole site these statistics :
A% victory with A or defeat with F ;
F% victory with F or ... ;
(1-A%-F% are draw) ;
G% victory with G or ...;
I % victory with I or ...

We just increase the ELO points that are exchanged in each game with these results :
* when A (resp. F, G, I) wins, player gets F%/50% = 2 * F% (resp. A%, I%, G%) of the usual elo points that are exchanged
* when A (reps. F, G, I) loses, player gets 2 * A% (resp F%, G%, I%) of the usual elo point that are exchanged.

That way, the countries that win often because they are unbalanced, win less and less points... And instead of arguing which country would be the best, we just measure it.

Re: May GhostRatings Published

by Mercy » Mon May 21, 2018 10:42 am

ghug wrote:
Mon May 21, 2018 6:31 am
I'm leaning towards this solution of a better combined ranking with the per-variant rankings listed in the spreadsheet, but more input is very welcome.
Since more input is welcome, let me give you my two cents. :)

I see two problems with combining FvA and GvI in one scoring system.
1) Your skill in FvA does not necessarily reflect your skill in GvI and vice versa.
2) The two variants are not equally balanced. In particular, GvI seems to be less balanced than FvA, meaning that in GvI, there is more luck involved in what country you get assigned. That means that if you care about your rankings, and you will play against a worse/lower ranked player, it is best for you to avoid GvI. If you will play against a better/higher ranked player, though, then it is better to take your chances and play GvI instead of FvA.

Problem 1) has been raised many times and it is impossible to solve in a ranking where both variants are combined. I don't mind problem 1) too much, though; the variants are similar enough that after some practice, your skill in one of them should be a pretty good indicator for your skill in the other.

Problem 2) hasn't been raised specifically, but I saw some mentioning about balance, so maybe some people have indirectly mentioned the problem. The good news is that problem 2) can be solved. I have thought a bit about ranking systems across different variants myself in the past and I have come up with the following.

The 1v1 ranking is based on ELO, right? That means that it is assumed that the performance of each player in each game is given by a random variable from a normal distribution, and that the player with the highest performance wins. The ranking of each player is then the best approximation of the mean of this normal distribution. If we assume that this all is the case, then the less balanced a variant is, the higher the standard deviation of this normal distribution should be to account for the extra luck involved in country assignment. Hence, increasing the standard deviation used for GvI games by the right amount will solve problem 2).

The right amount could in theory be calculated by the computer itself. The computer could look at all games played and for each variant determine how much the win percentage of a player is an indicator for it to win a game in that variant. The less this is an indicator, the less balanced a game is, and the higher the standard deviation should be.

I can imagine that that last part would be a lot of work to program, so we can be less precise and just try to guess ourselves what standard deviations should be assumed. A higher standard deviation should go hand in hand with a lower k-value, too.

Re: May GhostRatings Published

by CptMike » Mon May 21, 2018 7:23 am

MadMarx wrote:
Mon May 21, 2018 5:56 am
I suppose if you’ve played each variant over 100 times you likely have a different view of the situation, not sure how many people on webDip are in that club, and it may be the case that those that are more invested should have more say, though it may also be the case that GR should look to serve the 99.9% rather than the 0.1%, I honestly can’t say...
You can't say if promoting something that would serve only 0.1% rather than the 99.9% is good or not ? It is clear that your judgement is not very good then… Or maybe it is the statement that is not honest ? :eyeroll:

MadMarx, if there are some versions of the game that you don't want to play, why don't you just avoid playing them ? Feel free.

Instead you state that those who play them are 0.1% and suggest that taking them into account is "not overly valuable" ? "I honestly can't say" if this is respectful of them or not...
ghug wrote:
Mon May 21, 2018 6:31 am
I'm leaning towards this solution of a better combined ranking with the per-variant rankings listed in the spreadsheet, but more input is very welcome.
Ghug's proposal solves all the points that MadMarx put forward initialy : the new global GR based on the GR's of GvI and FvA, calculated separately. Being better at one or the other variant doesn't influence the result for the other variant any more.

Global GR is not just the sum of both divided by two. It weighs both individual GR by the number of games played by each player for each variant, meaning that if a player doesn't play many GvI or play them less often will not harm his global score at all... (eg FvA: 500 with 45 games and GvI 100 with 5 games would give a global GR of 460)

Re: May GhostRatings Published

by ghug » Mon May 21, 2018 6:31 am

I am willing to listen to opinions, which is why I wish more people would provide them :razz:

I'm leaning towards this solution of a better combined ranking with the per-variant rankings listed in the spreadsheet, but more input is very welcome.

Re: May GhostRatings Published

by MadMarx » Mon May 21, 2018 5:56 am

CptMike wrote:
Mon May 21, 2018 2:45 am
It is of course valuable to be able to play both versions.
I suppose if you’ve played each variant over 100 times you likely have a different view of the situation, not sure how many people on webDip are in that club, and it may be the case that those that are more invested should have more say, though it may also be the case that GR should look to serve the 99.9% rather than the 0.1%, I honestly can’t say...

In the end it’s pretty cool to have any GR calculated on a regular basis, and it’s also cool that GR2 is around to provide alternate rankings/categories, you ultimately can’t go wrong whatever you decide, especially if you’re willing to take the time to listen to our opinions and consider them, thank you, much appreciated!

Top