Massive Assault

New Glicko Ratings Page
Page 1 of 4

Author:  Maelstrom [ Mon May 24, 2004 1:09 pm ]
Post subject:  New Glicko Ratings Page

The player Norat is a good php web developer, and has offered to help redesign the Glicko Ratings page. It has had a brand new facelift, and it looks great! In the process, he also found some minor errors with the Glicko functions that were used in the old site.

Here's the link:

Some new features of this page:
1) All data is pulled directly from the MA database, so it is always up to date

2) We now have a date when each game is completed, so the problem where we rated games by when they were started rather than when they were finished is now fixed.

3) There is an added feature to filter the list by an RD of your choice.

4) We've got national flags!

Norat has also opened up new possibilities of how we can sort the data.

Author:  Artanis [ Mon May 24, 2004 1:48 pm ]
Post subject: 

Sweet! *races for the "bookmarks" button on his browser*

Author:  Three Seven [ Mon May 24, 2004 2:52 pm ]
Post subject: 


Author:  Groucho [ Mon May 24, 2004 2:54 pm ]
Post subject: 

Beautiful. Very nice.

Now, if only this can be incorporated into the game itself.

Author:  Maelstrom [ Mon May 24, 2004 3:07 pm ]
Post subject: 

This new page lets us continue to advance our rating system. One option that has been left out so far is the option of taking into account time in ratings.

Point 1) Here is a problem with the current ratings, presented by our marvelous php developer in an email to me:
How it calculates now:
"Rating interval" equals zero. Ratings are updated on game-by-game
basis. The only check is that RD never exceeds 350.
RD' = min(350, RD).
Example for problem: there are absolutely equal players (ratings and
RDs). They play 2 simultaneous games (common case: same map, different
sides). First player wins one game and second player wins another
one. It would be ideal if players ratings aren't changed (because
players equal), only RDs changed. But now when second game is
calculated players are not equal (because of first game) and final
ratings will be inequal.
I think this is minor problem and it should be discussed in forum.
As you can see in gdescrip.pdf FICS uses game-by-game system also (as
we now).
The question is Is it really neccessary and how long should interval
be: one week, less, more? Note that Glicko works best if there are not
more than 5-10 games in rating interval per player.
Bad news: in above example if players finish games on interval's
border, ratings will be inequal again. But this situation will be

I personally don't think this is a big deal, and a tough one to solve in the case of MAN. Glicko is based on a chess system, but chess games are finished in one sitting. MAN games can take anywhere from a few hours to a couple months, depending on the availability of the players. Because of this I don't think that we should rate games based on time.

Point 2) Another problem we have though is inactive players. One solution presented by the Glicko system is to steadily increase inactive players' RD over time, so in effect we are saying we don't know how accurate the scores are anymore. I personally think this is a great idea.

Point 3) Along with this, we can have an option to sort by "pessimistic rating", which is your rating minus your RD. That puts players that have played more and played well near the top, with those that have played less but played even better slightly lower.

Any thoughts on these items?

Author:  Artanis [ Mon May 24, 2004 3:17 pm ]
Post subject: 

Points 2 and 3 look good

About the simultaneous games thing: how about if two games on the same map between the same players are started within a game number difference of each other (say up to 5 games apart, i.e. 9941 and 9943 would be within this barrier) then they're taken as a pair for the system: if one player wins both, then he's credited for two wins, but if they split, they're both credited for two draws.

Author:  Vixen [ Tue May 25, 2004 3:31 am ]
Post subject: 


What about sorting by RD and then by Rate when was choosed sort by RD?

Notes: is there the small misprint? - "Hide players with RD less than"

Author:  redfox [ Tue May 25, 2004 1:34 pm ]
Post subject: 

For suggestion #3, would this pessimistic ratings be a variation on the official ratings, or would it become the new default ratings? I be interested to see this variation of the data, but I also think the regular view is fine as well. A filtering of RD<60 seems to give an accurate depiction of the top experienced players, though there are always a few outliers.

Suggestions #2 sounds good as well. I don't think there's too much we can do about #1, so we can do it either way.

Author:  Maelstrom [ Tue May 25, 2004 3:41 pm ]
Post subject: 

redfox wrote:
For suggestion #3, would this pessimistic ratings be a variation on the official ratings, or would it become the new default ratings? I be interested to see this variation of the data, but I also think the regular view is fine as well.

It'll be an additional option for sure. I'm sure new people like the idea of stacking up to the veterans at least temporarily ;)

Author:  redfox [ Tue May 25, 2004 4:22 pm ]
Post subject: 

heh, well, if anyone with skill wanted to be at the top of the list, they could just start a new account and beat a bunch of noobs, that way their score would only go up.

In fact, if you were really desperate, you could do that with one account (Account A), get it to a high score, then create another account (Account B), and use it to beat Account A a bunch of times. Unfortunately, this is something Glicko is not meant to take into account, but I hope no one is desperate enough to stoop to that level.

Is there a way to filter out Free accounts (aside from taking out results from Emerald, which might not be a bad idea either)?

Author:  Norat [ Tue May 25, 2004 6:06 pm ]
Post subject: 

To Vixen: the RDs are sorted properly. The point is that Ratings and RDs are floating point numbers, but on the page they are just rounded. This can result in something like:<pre>Rating RD
1750 350
1800 350</PRE>but actually there are:<PRE>1750.23 349.78
1800.56 350.12</PRE>and RDs are not equal.

Misprint fixed. Thank you.

Author:  Quitch [ Wed May 26, 2004 4:54 am ]
Post subject: 

Looks pretty sweet.... I just dropped 40 places! Youch.

The "less than" filter is in actual fact a "less than or equal to" filter, this should be corrected (or noted).

Author:  Artanis [ Thu May 27, 2004 5:41 pm ]
Post subject: 

Something I just thought of: would it be possible to count a player's regular and MENTOR accounts as one person?

Author:  Norat [ Thu May 27, 2004 6:45 pm ]
Post subject: 

It is possible. But do our Mentors want this? I also need a list with mentor and corresponding normal names.

Author:  Quitch [ Fri May 28, 2004 2:37 am ]
Post subject: 

I assume at the moment that ranking changes are figured out via the main ranking? Since chess has only one board and MA several, wouldn't it make more sense to work out the score by planet, then average the scores and RDs to give you the overall score and RD? Thus, beating Redfox on Bizarria would give you the boost you deserve, while whooping someone on a map their ranking shows they suck at, wouldn't affect your score much (e.g. Someone is good at every map except Wasserland, has a high score and low RD, so obviously you challenge them at Wasserland for an easy rankings boost... this system would prevent that).

Author:  Norat [ Fri May 28, 2004 4:26 am ]
Post subject: 

IMHO there is no point, because if you are good, you are good on every map, and if you play badly, you play badly on every map. It seems to me that your MA skill almost doesn't depend on map (well it may depend on map size slightly). I think the reason someone good has low rating on some map is that he just played few battles on it. If he would play more battles, he will have high rating on this map also.
And another problem: trial players. They can play only on Emerald. But there are 9 maps! Rating = (8*1500+very_high_rating)/9 ? So the trial players can not be on top by design :( The stimulus to subscribe? ;)
It will be funny to look how you will "boost" your rating by challenging Tiger on Crateus :lol:

Author:  Quitch [ Fri May 28, 2004 7:22 am ]
Post subject: 

I don't think that's true. I know for example that mrakobes can beat Tiger on some maps well enough, but on Wasserland and Anubis always loses. Redfox looks nigh undefeatable on Bizzaria, but more vunerable elsewhere.

It's not a case of one global tactic for all maps, and thus the system should reflect that. I'd agree with you if the per planet list reflected the global list, but it doesn't so your point cannot be true.

Author:  Maelstrom [ Fri May 28, 2004 11:06 am ]
Post subject: 

I agree with Quitch that each map has its own style, but I don't like equally weighting them in an overall weighting system. There are some people that are especially good at certain maps. For some reason, I've never been able to get the hang of Anubis, while I hold my own on Antarticus or Wasserland.

By forcing the ratings to be equally divided among all maps though is extreme. This assumes that all maps are equal in strategy, and assumes that to be the best player, you must master all levels of strategy and tactics found in the different maps. It really hurts though if you specialize on certain maps for one reason or another.

I think this would be an interesting ranking, to see who is best all around, but I don't think it should be the main system. It can be an additional system. It may also be best to weight the maps, but then again you're assuming that mastering one aspect of the game is better then mastering another.

If we were going to do it right, and fully specialize this system for the unique qualities of Massive Assault, we should also include which side was played in each game. This would be weighted heavier on the smaller maps, so that if you pull off a victory against the odds of having FNU on Emerald against a good opponent, you should be rewarded for that.

Author:  Quitch [ Sat May 29, 2004 4:15 pm ]
Post subject: 

Introducing weighting is a bad idea. It's taking a heavily tested, proven statistically accurate system, then shoving a random x2 in there. Why bother? We can do it by map, we can do it by side, and for an overall rating we can average it out. The overall rating HAS to be the main rating because people always want a single figure.

In my world your glicko rating changes as per normal, but takes into account the map and the sides (so if you play as PL on Emerald the score would be worked out using your PL on Emerald rating & RD against their FNU on Emerald rating & RD.) Once your new score and RD are updated, a new average glicko rating is calculated for your overall rating and RD. I don't see a better, Glicko way of doing it. Not only does this take map into account, but it also compensates for the FNU being tough on small maps.

Glicko would now break down into the following tables

Overall (your average score and RD from every map & side)
Emerald PL
Emerald FNU

This would give you a wide variety of information AND lead to far more accurate scoring, taking better into the account the new factors that MA introduces.

Author:  Maelstrom [ Sat May 29, 2004 4:53 pm ]
Post subject: 

Hmm, a very good approach. I like the idea of having a seperate FNU and PL rating, as that removes the need to try to come up with a way to accurately weight them. This would be done automatically, as your PL rating will ususally be higher than your FNU rating. These two scores could then be averaged for that specific map to give you your map rating.

I'm still not comfortable with averaging the seperate maps together, as that still assumes all maps are equal in the skill they show you have. I think we're going to just have to agree to disagree on this point ;)

I'd like to hear some input from more players on this. Once we've made the decision on how to do that, Norat will apply that to the Glicko system.

So the question is to other players:
Should the maps be weighted equally as Quitch proposes?
Or should they each have a seperate weight? If so, how would you weight them?

Of course, we may be able to do it both ways, that way everyone is happy ;)

Page 1 of 4 All times are UTC - 5 hours
Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group