Forged Alliance Forever

Re: Server update today (14/1-2015)

2015-01-22T22:10:24+02:00

Right, just so we don't drop the ball here.

If someone can formulate these ideas into code, they should be looking at the following function in ladderGamesContainer: https://github.com/FAForever/server/blob/develop/gameModes/ladderGamesContainer.py#L141

Statistics: Posted by Sheeo — 22 Jan 2015, 22:10

Re: Server update today (14/1-2015)

2015-01-21T17:23:36+02:00

Just wanted to remind that current system has flaws.
Yesterday last 2 games i played were on Twin Rivers.
Aaaaand the next 4 maps i got today were..... Twin Rivers too!
That makes it 6 Twin Rivers in a row. You gotta be shitting me

Statistics: Posted by Mad`Mozart — 21 Jan 2015, 17:23

Re: Server update today (14/1-2015)

2015-01-17T20:22:04+02:00

Another thing to be careful of, but if players only have 2 or 3 maps in common and they get the same players over and over again in ladder, they are going to get the same maps over and over again.

Statistics: Posted by Kalvirox — 17 Jan 2015, 20:22

Re: Server update today (14/1-2015)

2015-01-17T17:45:04+02:00

but I think the objective should be to have a map both players want to play more often than straight chance, so instead of a 50/50 split, I still think there should be a greater bias towards maps both players have selected. Not too much bias though, or you end up in the situation which began this thread where a player rarely sees their other maps when playing against the same guy over and over again.

You can try to modify it to allow that, but it is tricky. It is hard to keep it fair. The maps that a player has selected is his ideal ladder pool. By selecting a set of maps he is saying: I would like to play with that pool. For example if a player has selected all maps, what that means is that he would like to play a totally random map, because he likes diversity. However, suppose that his opponent has selected only 5x5 maps. So the shared maps are the 5x5 maps. If you give extra bias to the shared maps, that would make the first player unhappier: he wants to play a diverse set of maps. It would make his opponent happier, because he gets to play more 5x5 maps. So if you bias it in a simple way, then it is unfair to people who select a lot of maps.

So with biasing toward shared maps, you have to be careful to only do it when it would make *both* players happier, and not do it when it would make one player happier and one player unhappier. For example if one player has selected all 5x5 and 10x10 maps, and the other has selected all 10x10 and 20x20 maps, it might make sense to bias toward 10x10 maps, because it makes *both* players happier (of course you still have to be careful not to bias too much, to make sure that you don't get the problem where you get the same maps over and over again, and almost never the maps that are not shared).

Mathematically you have to take care that if you bias in such a way to increase the chances by x% that the map that is picked is in player 1's selection, then you must also increase by x% the chances that the map that is picked is in player 2's selection. It's like a trade: player 1 says, ok, because I don't want to play that much on 20x20 maps, in return I'm willing to not play that much on 5x5 maps.

How you would do that is a little bit difficult to explain, but I can code it if people want this.

Statistics: Posted by Vee — 17 Jan 2015, 17:45

Re: Server update today (14/1-2015)

2015-01-17T00:10:27+02:00

So I have no maps selected, how does that affect what maps i get.

Statistics: Posted by Aulex — 17 Jan 2015, 00:10

Re: Server update today (14/1-2015)

2015-01-16T21:48:49+02:00

Love how this topic has moved forwards Just a few things to say.

First, is it really SO bad to force players to select 15 maps? Is it really so bad, that a player who wants to play ladder, step one of that world is 'Select 15 maps you would prefer to play'? I think not.

Second, Vee's idea is really great, but I think the objective should be to have a map both players want to play more often than straight chance, so instead of a 50/50 split, I still think there should be a greater bias towards maps both players have selected. Not too much bias though, or you end up in the situation which began this thread where a player rarely sees their other maps when playing against the same guy over and over again.

Third, no matter the system used, I think it would be a good idea to have the probability of a particular map being selected reduce each time it's picked, and reset when it's not, to reduce the chances of seeing it too often. Human perception of probability counts almost as much as true probability in this discussion, because it's about what annoys humans

Statistics: Posted by IceDreamer — 16 Jan 2015, 21:48

Re: Server update today (14/1-2015)

2015-01-16T14:27:15+02:00

Sir Prize wrote:
Second edit: I think this has kind of been said but if between two players there are 15 unique maps selected, the popular pool shouldn't be referenced imo.

It won't.

Edit:

Sir Prize wrote:
Edit: I think biasing towards the shared map pool is a good thing for the vast majority of players, but is it possible to weight the probability away from maps you've recently played or recently played multiple times Sheeo/Zep? Perhaps optionally? I personally have no problem getting a map I like repeatedly, Mozart obviously is over it, which is fine and I'm sure theres plenty of people that would feel like either of us.

It's an option yeah, but it should not be necessary overall. I'm against manipulating uniformly random distributions because you often end up making them too biased

Statistics: Posted by Sheeo — 16 Jan 2015, 14:27

Re: Server update today (14/1-2015)

2015-01-16T13:48:15+02:00

Might have got a bit bitchy but this is a really good discussion, lots of good ideas. I think doing popular maps by ladder rating bracket would be really good. I also still like the idea of giving players on opportunity to veto a map and make the system pick a new one when two players are matched, perhaps if they have just played a map 3 times in their last 7 games or something. Ideally the system would remember the veto for a bit as well, dunno if that's viable and I haven't thought it through.

Edit: I think biasing towards the shared map pool is a good thing for the vast majority of players, but is it possible to weight the probability away from maps you've recently played or recently played multiple times Sheeo/Zep? Perhaps optionally? I personally have no problem getting a map I like repeatedly, Mozart obviously is over it, which is fine and I'm sure theres plenty of people that would feel like either of us.

Second edit: I think this has kind of been said but if between two players there are 15 unique maps selected, the popular pool shouldn't be referenced imo.

Statistics: Posted by Sir Prize — 16 Jan 2015, 13:48

Re: Server update today (14/1-2015)

2015-01-16T11:08:16+02:00

Indeed ladder rating is more natural to use, I just happened to type out global_rating when doing the queries.

Instead of static ranges, another possibility is to use [min{p1,p2}-some_constant, max{p1,p2}+some_constant]

Statistics: Posted by Sheeo — 16 Jan 2015, 11:08

Re: Server update today (14/1-2015)

2015-01-16T11:03:17+02:00

Hm, if talking about such brackets for ladder, it would be only natural to use ladder rating.
I would suggest something like:

<1000
1000-1300
1300-1600
1600-1900
>1900

This way bottom and top of the bracket are in the search range.
Or could take some other step, like 350 or 400 to make it more simple.

Statistics: Posted by Mad`Mozart — 16 Jan 2015, 11:03

Re: Server update today (14/1-2015)

2015-01-16T11:02:48+02:00

Vee wrote:
I do think it's safe to assume a fair coin from the standard library of python and a normal operating system.

Sure, but subjective fairness as perceived by humans does not always follow the laws of probability For example it's quite possible to get 4 maps in a row from your opponent, since that's 1/2^4 = 12.5% chance, but most people would perceive that as unfair. Especially when a game takes like 30 minutes each, then when the get a map from their enemy for the 4th time in a row they may think "WTF again??".

Thanks for the map statistics! Those common maps for the top players are a lot more fun & varied than the common maps for the entire userbase, IMO.

Sure, I'm not convinced that this is worth implementing, however, since there are relatively small chances of perceived 'unfairness' happening relative to 'fairness'

No problem! If you want any more specific stats (I didn't have the time to investigate more just now) please let me know.

Statistics: Posted by Sheeo — 16 Jan 2015, 11:02

Re: Server update today (14/1-2015)

2015-01-16T11:08:11+02:00

I do think it's safe to assume a fair coin from the standard library of python and a normal operating system.

Sure, but subjective fairness as perceived by humans does not always follow the laws of probability For example it's quite possible to get 4 maps in a row from your opponent, since that's 1/2^4 = 12.5% chance, but most people would perceive that as unfair. Especially when a game takes like 30 minutes each, then when the get a map from their enemy for the 4th time in a row they may think "WTF again??".

Thanks for the map statistics!

Here are the maps in common for the >2k and <1k:

Open palms
Twin rivers
Badlands
Williamson's bridge
Canis river
Blasted rock
Fields of isis
Finn's revenge
Ambush pass

Maps for >2k:

Theta passage
Loki
Syrtis major
Eye of the storm
Desert arena
TAG craftious maximus

Maps for <1k:

Winter duel
Balvery mountains v2
Seton's clutch
Crag dunes
Waters of isis
Four leaf clover

Statistics: Posted by Vee — 16 Jan 2015, 10:54

Re: Server update today (14/1-2015)

2015-01-16T10:47:25+02:00

Vee wrote:
Since the worst maps were removed, there is not really a reason any more to bias so much to the commonly selected maps, and to the maps that players have in common. I think the system can simply work like this:

When picking a map, flip a coin, and with 50% chances pick a random map from player 1's selected maps, and with 50% chances pick a random map from player 2's selected maps. When a player has less than 15 maps, fill it up to 15 with common maps or random maps. It would be nice if the common maps were based on brackets, so that the "common maps" for the top 50 ladder don't get totally decided by the 2000 players below that.

If necessary to make it more fair, you can keep track of the count that the coin flip went in a player's disadvantage, minus the count of the coin flips that went in a player's advantage. So if your count is 3, it could mean for example that 13 times you won the coin flip, but 10 times one of your opponents won the coin flip. With that number for both players, you could bias the coin to more or less than 50% depending on who has the higher number. For example if the numbers are n and k, then you could take probability proportional to 2^n for player 1 and proportional to 2^k for player 2 instead of 0.5. So the total probability would then be 2^n/S for player 1, and 2^k/S for player 2, with S = 2^n + 2^k. Or you could simply take probability is 0.5 if n=k, probability is 0 if n>k and probability is 1 if n

Very interesting. I do think it's safe to assume a fair coin from the standard library of python and a normal operating system.

Regarding the bracketing of ladder map selection, that could be very interesting to investigate.

This should be the top 15 for mean > 2k (Global rating):

Code:: mysql> select idMap, count(idUser) as amount, name from ladder_map_selection inner join table_map on idMap = table_map.id inner join global_rating on global_rating.id=ladder_map_selection.idUser where global_rating.mean > 2000 group by idMap order by amount desc limit 15; +-------+--------+-----------------------+ | idMap | amount | name | +-------+--------+-----------------------+ | 558 | 50 | Open Palms | | 156 | 47 | Twin Rivers | | 5 | 47 | 8 - Badlands_v4 | | 563 | 44 | Theta Passage | | 8 | 43 | loki | | 568 | 43 | Syrtis Major | | 566 | 40 | Fields of Isis | | 605 | 39 | Eye of the Storm | | 567 | 38 | Canis River | | 587 | 37 | Blasted Rock | | 593 | 37 | Williamson's Bridge | | 570 | 36 | Finn's Revenge | | 267 | 34 | desert arena | | 589 | 34 | Ambush Pass | | 88 | 32 | TAG_Craftious Maximus | +-------+--------+-----------------------+ 15 rows in set (0.14 sec)

Top 15 for mean < 1k (Global rating):

Code:: mysql> select idMap, count(idUser) as amount, name from ladder_map_selection inner join table_map on idMap = table_map.id inner join global_rating on global_rating.id=ladder_map_selection.idUser where global_rating.mean < 1000 group by idMap order by amount desc limit 15; +-------+--------+----------------------+ | idMap | amount | name | +-------+--------+----------------------+ | 566 | 507 | Fields of Isis | | 570 | 488 | Finn's Revenge | | 567 | 461 | Canis River | | 587 | 434 | Blasted Rock | | 564 | 433 | Winter Duel | | 589 | 427 | Ambush Pass | | 264 | 425 | Balvery Mountains V2 | | 5 | 405 | 8 - Badlands_v4 | | 560 | 395 | Seton's Clutch | | 592 | 381 | Crag Dunes | | 593 | 371 | Williamson's Bridge | | 558 | 356 | Open Palms | | 57 | 355 | Waters of Isis r001 | | 156 | 350 | Twin Rivers | | 582 | 345 | Four-Leaf Clover | +-------+--------+----------------------+ 15 rows in set (0.13 sec)

Statistics: Posted by Sheeo — 16 Jan 2015, 10:47

Re: Server update today (14/1-2015)

2015-01-16T10:36:07+02:00

Sheeo wrote:
Actually there have been more than that over the past few months, but I'm not going to satisfy your need for hostile discussion further.

Ah yes, if you think im trying to blame you here, you got the wrong idea.
Not today and not in this thread, you're not even related. But i promise to try to find something in the future

Statistics: Posted by Mad`Mozart — 16 Jan 2015, 10:36

Re: Server update today (14/1-2015)

2015-01-16T10:44:38+02:00

Since the worst maps were removed, there is not really a reason any more to bias so much to the commonly selected maps, and to the maps that players have in common. I think the system can simply work like this:

When picking a map, flip a coin, and with 50% chances pick a random map from player 1's selected maps, and with 50% chances pick a random map from player 2's selected maps. When a player has less than 15 maps, fill it up to 15 with common maps or random maps. It would be nice if the common maps were based on brackets, so that the "common maps" for the top 50 ladder don't get totally decided by the 2000 players below that.

That way the maps that both players have in common have twice the chance to be played, but you still get map diversity because even the maps that they don't have in common will be played a lot.

If necessary to make it more fair, you can keep track of the count that the coin flip went in a player's disadvantage, minus the count of the coin flips that went in a player's advantage. So if your count is 3, it could mean for example that 13 times you won the coin flip, but 10 times one of your opponents won the coin flip. With that number for both players, you could bias the coin to more or less than 50% depending on who has the higher number. For example if the numbers are n and k, then you could take probability proportional to 2^n for player 1 and proportional to 2^k for player 2 instead of 0.5. So the total probability would then be 2^n/S for player 1, and 2^k/S for player 2, with S = 2^n + 2^k. Or you could simply take probability is 0.5 if n=k, probability is 0 if n>k and probability is 1 if nStatistics: Posted by Vee — 16 Jan 2015, 10:32