Often getting the same map in ladder

[Vee]

It is a well known problem that you get the same maps over an over again in ladder, while you never or almost never get some maps that you have selected. Which map you get all the time depends on you and your opponent. Some people get badlands all the time, some emerald crater, some winter duel. Maybe people are just imagining that and it's truly just randomness...but it's still worth to investigate. I've looked at the code here https://github.com/FAForever/server/blo ... er.py#L130 This code seems OK. Perhaps the problem is that the table ladder_map_selection contains duplicate entries, so that you have a higher chance to get those maps that have duplicate entries. Or perhaps it is the algorithm after all. How about changing it to this:

Code: Select all

def choose_ladder_map_pool(self, player1, player2):
   lucky_player = [player1,player2][random.randint(0,1)]
   maps = self.getSelectedLadderMaps(lucky_player.id)
   missing_maps = 15 - len(maps)
   if missing_maps > 0:
      maps += self.getPopularLadderMaps(missing_maps)[:missing_maps]
   return maps

Explanation: we randomly pick one of the two players's map pools. If that map pool has less than 15 maps, we extend it to 15 with popular maps. Then we pick a random map from that list. This code is simpler and will result in a more diverse set of maps being played because all maps that a player has selected have some chance to be picked, rather than just the intersection. Also the 33% popular maps chance has been removed. Now you will only get your mappool extended with popular maps if you have selected less than 15.

[ZLO_RD]

how do extend someone's pool? are you sure you extend with random maps in your algorithm? (i am not good at coding i guess)

Edit: i kinda like idea, not sure about removal of common maps thought.

[Vee]

The pool gets extended in the same way as it gets extended now: by adding popular maps. So if you have 9 maps selected then it will add the 6 most popular ladder maps to that. So it is not extended by random maps. That would be easy to change though. See the code here: https://github.com/FAForever/server/blo ... er.py#L109

Common maps still have more chance to be played because they are in the pool of both players so they have double chance to be played, but now you will not only get common maps. We could also give some extra chance to common maps to make it more than double, just not 100% chance. But if the database contains duplicate rows then that should be fixed first of course (that is the only other reason I can think of why some people often get the same map).

[Sheeo]

Perhaps the problem is that the table ladder_map_selection contains duplicate entries

This assumption is false: https://github.com/FAForever/server/blob/develop/src/lobbyconnection.py#L1054

While the code is a bit contrived, duplicates are indeed filtered on insertion to the database.


I could analyze the map selections of some players who claim they are unfairly given the same maps, if you'd like.

[speed2]

What about some code that would not allow to pick same map again. For example if you played one map, you wont get it for another lets say 5 games (maybe more). Cause even I played a bit ladder and got 3x in a row badlands...and that made me quit the last one cause thats just dumb.
How hard would be this change?

[Sheeo]

So difficult that the following query would do it

Code: Select all

(SELECT mapid
FROM   game_stats
        INNER JOIN game_player_stats
                ON game_player_stats.gameid = game_stats.id
WHERE  playerid = ?
LIMIT  5)
UNION DISTINCT
(SELECT mapid
FROM   game_stats
        INNER JOIN game_player_stats
                ON game_player_stats.gameid = game_stats.id
WHERE  playerid = ?
LIMIT  5);


Having two select clauses ensures we get the latest 5 maps for each player.

I'm assuming this isn't very controversial, so I'll get this into the network update.

[Vee]

Thanks that is great! That solves the issue that you get maps multiple times in a row. There is still the other issue that you (almost) never gets some maps that you have selected. Or maybe we are just imagining that, humans are not very good at probability by feeling

[ZLO_RD]

Give human random data, and he will find tons of patterns in it

i actually heared story that some one replaced random algorithm to a not random one cause users were compainig that it was not random enought...

[Gorton]

I wouldn't say 5 games, but at least 2-3.

[Vee]

Maybe we are not imagining it after all (from TheRedViper)

p < 10^-graham's number