ELO corruption

TalonZa
Script
Script
Posts: 24

ELO corruption

Post#1 » 31 Jul 2017, 17:55

There is a problem with the ELO system and this game given that you queue up 10 games. When you return later you find that you were matched up to the same opponent several times on the same map. The games are exactly the same yet you looose like 60 ELO.

There needs to be a fix. Perhaps prevent consecutive maps against same opponent on the same map over a 10 game cycle.

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#2 » 31 Jul 2017, 18:21

I disagree for the following reason.

1. first if you lose against an opponent, there is little to complain about, 60 elo or 300, he deserved to win those points.

2. if you queue the current cool down is of 5 minutes, so to be "picked" by the same player, say, 5 times, the other player has to play at least different times in 25 minutes. I think it is ok. Otherwise you may wait hours or days for different players and I think the game is slow enough that further slow down will be disliked.

3. The playerbase is small in master and grand master league, there is little to do (unless you go around and advertise the game), there are always the same 4-5 players. I already read 2 or 3 complains like this from you, for my point of view it means: you need to improve against those players, not hoping to gain points just spamming games.


To make it short: if one hopes to improve his score avoiding certain opponents, it is cheap. The score, within 100 points of tolerance (in master and grand master, in other leagues this is larger due to larger kfactor), is telling properly the strength of a player compared to the others. So trying to get "higher" without the skill to resist to certain players, is preparing the ground for large falls. Like you said, 60 points.

I remember one day I went from 2030 to 1890 because the previous day I was lucky with maps and opponents. Luck, on long term, does not help in this game. Either one is strong, or one is strong, there is no other way.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 563

Re: ELO corruption

Post#3 » 31 Jul 2017, 19:36

pier4r wrote:I disagree for the following reason.

1. first if you lose against an opponent, there is little to complain about, 60 elo or 300, he deserved to win those points.


I agree with you pier but i understand so much TalonZa.
You cant see how your improvement/modification of you AI works if you play always against the same opponent on the same map. I want to see how the modifications react to multiple AI not only 1 or 2


pier4r wrote:3. The playerbase is small in master and grand master league, there is little to do (unless you go around and advertise the game), there are always the same 4-5 players. I already read 2 or 3 complains like this from you, for my point of view it means: you need to improve against those players, not hoping to gain points just spamming games.


During this period (i mean new leagues and matching systems), I think we will lose more player than new players will come (Me i finish the tournament i'm involved to and i stop)


I think, due to the lack of player, that we should think to another system of ranked.

You will says that we have freeplay to play as we want (maybe some guys will do that but i dont wanna as i prefer play for rank when i can)

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#4 » 31 Jul 2017, 19:55

nah freeplay is a meh. Aside from tournaments, there are little players.

And even if you bring 4 million new players, if they are easy and stick to the game only 40 games, nothing is solved for the top leagues.

because:
(a) people normally play games that do not require much brain effort for long periods. So the game may attract them for a few games and that's it. In the words of my brother "nah, if I have to spend hours on it, I prefer to study".

(b) if a player is not that strong, what is his value for the master / grand master players? Mostly cannon fodder plus an occasional upset. So do you want easy wins and a lot of +1? Nah.

(c) due to a and b the valuable information comes from players of the same skill or above, and the amount of players is a little. Either one deal with it (for me a game can be interesting as long as there is one person that I have to beat) or, well, there is little to do. Unless one takes the duty (that is a job) to advertise the game to collect those 1 out of 1000 players that stick to the game for over 1000 games.



For me, in grandmaster (but the same would be in master league) I focus to not lose __at all__ to players under me or equal to me, even the same person. It is easier against the same person because I have a lot of data of him.

The result of this is of course that my score raises so I have more opponents under me, making the job more difficult.

So for example I'll focus to avoid this: well I was checking the last matches of talonza but he loses mostly to equal score players, nothing bad. The score is indeed working. Almost equal score means = 50% chances of winning/drawing on average.

Same for you Kid. You seems losing against equal score guys. Either you improve over them or, well, it will be the same. But it is also true that only assault is already great to keep floating around 1700.

You should start using other classes.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 563

Re: ELO corruption

Post#5 » 31 Jul 2017, 20:25

I am not complaining abotu my elo, but the variation o opponent has drastically fall and i found that not cool (I even have the idea to voluntarily lose to see if there are more people in platinum).

And for my score, my latest change does not work that good against 0x1de, Arcueid-7 and TalonZa (whose represent 90% of master league), i would like to play against more player (even if they are better, i do not care about their rank).

i was thinking about a system of only 2-3 leagues (néophytes-vétérans-pros) and why not with promotion/demotion by week

pier4r wrote:It is easier against the same person because I have a lot of data of him.

Super; you are just managing to beat specific people (i do not consider that as improving your AI)

Arcueid_57
Script
Script
Posts: 41

Re: ELO corruption

Post#6 » 31 Jul 2017, 20:33

I posted a proposal here:
viewtopic.php?f=2&t=902

It won't solve the problem but it should improve the situation at least in Master League.

If many people post a "+1" maybe it will be implemented.

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#7 » 31 Jul 2017, 21:20

TheKidPunisher wrote:Super; you are just managing to beat specific people (i do not consider that as improving your AI)


I manage to optimize against the active playerbase, as in every game players do.

Otherwise if I have to optimize against all the possible enemy ais, I could play 500k games without even scratch the surface of possibilities, at least I think so.

It is like I say to you: go an write down all the possible sums of integer numbers that results in 100'000, by hand.

As in 99'999 + 1 ; 99'998 + 1 + 1 ; 50'000 + 50'000 , etc.

Good luck with that.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 563

Re: ELO corruption

Post#8 » 31 Jul 2017, 23:09

pier4r wrote:Otherwise if I have to optimize against all the possible enemy ais,


That's what i call improvement, and it is not by playing against 3-4 players that i will see the result


PS (no offense):
I understand you what you want to say.
But take a bigger number to correspond to the «without even scratch the surface of possibilities»

TalonZa
Script
Script
Posts: 24

Re: ELO corruption

Post#9 » 31 Jul 2017, 23:43

pier4r wrote:1. first if you lose against an opponent, there is little to complain about, 60 elo or 300, he deserved to win those points.


This is where I disagree... I have had matches like this:

Player 1 | Map 1 | Lose = -18
Player 1 | Map 1 | Lose = -17
Player 1 | Map 2 | Win= +17
Player 1 | Map 1 | Lose = -16
Player 1 | Map 3 | Win = +15
Player 1 | Map 4 | Win = +16
Player 1 | Map 1 | Lose = -16

All 4 those bold games were identical. Basically a replay of each other. So I lose 67 points because of 1 player on 1 map. Yet I beat the player on all other maps. It shouldn't be allowed.

The problem is made worse by so few players:
1 You're likely to play the same guy in consecutive matches
2 There not enough players to start games, so if you want to see results, you have to queue up 10 games

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#10 » 31 Jul 2017, 23:52

TheKidPunisher wrote:
pier4r wrote:Otherwise if I have to optimize against all the possible enemy ais,


That's what i call improvement, and it is not by playing against 3-4 players that i will see the result



You cannot optimize against something that you don't see. Either someone discovers it for you, and you are against it, or you have to discover it and it is very difficult, so your point is a bit fictional.

Moreover, if the AI should work against "all the possible others" all the more reason that it has to work against the existing players.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#11 » 31 Jul 2017, 23:55

TalonZa wrote:
pier4r wrote:1. first if you lose against an opponent, there is little to complain about, 60 elo or 300, he deserved to win those points.


This is where I disagree... I have had matches like this:

Player 1 | Map 1 | Lose = -18
Player 1 | Map 1 | Lose = -17
Player 1 | Map 2 | Win= +17
Player 1 | Map 1 | Lose = -16
Player 1 | Map 3 | Win = +15
Player 1 | Map 4 | Win = +16
Player 1 | Map 1 | Lose = -16

All 4 those bold games were identical. Basically a replay of each other. So I lose 67 points because of 1 player on 1 map. Yet I beat the player on all other maps. It shouldn't be allowed.

The problem is made worse by so few players:
1 You're likely to play the same guy in consecutive matches
2 There not enough players to start games, so if you want to see results, you have to queue up 10 games


Well you can adapt the deployment as I do, since you control it.

Do you know that there are mostly the same players? Yes.
Could you control the deployment? Yes

Knowing this, I deploy until I get the same map twice. So once I deploy 2 matches, another time 6, another 8 and so on. Plus I "use up" the available matches that were waiting, so I normally do runs of 10 games.

Plus, about losing points:
Ranked [2017-07-31 23:47:22] 742761 sollniss (1909) 2070 -14 Circle of Death
Ranked [2017-07-31 23:46:52] 742771 DH! (1934) 2064 +6 Barred Spiral (community)
Ranked [2017-07-31 23:38:14] 742785 Ritter Runkel (2096) 2073 -9 Kingmaker (community)
Ranked [2017-07-31 23:37:24] 742788 Milkhunter (2150) 2060 +13 Meeting point
Ranked [2017-07-31 23:30:48] 742759 sollniss (1909) 2074 -14 Split team
Ranked [2017-07-31 23:17:56] 742767 DH! (1941) 2088 -14 The seven wonders
Ranked [2017-07-30 23:41:17] 742083 Ritter Runkel (2128) 2107 -9 Set your priorities
Ranked [2017-07-30 23:34:52] 742082 Ritter Runkel (2130) 2107 -9 Meeting point
Ranked [2017-07-30 23:17:34] 742081 Milkhunter (2079) 2107 +0 Split team
Ranked [2017-07-30 23:15:11] 742080 LuBeNo (2073) 2107 -11 Circle of Death

It happens, nothing bad, all data (except cod, I refuse to fix that map for a while)
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 563

Re: ELO corruption

Post#12 » 01 Aug 2017, 00:28

pier4r wrote:
You cannot optimize against something that you don't see. Either someone discovers it for you, and you are against it, or you have to discover it and it is very difficult, so your point is a bit fictional.

Moreover, if the AI should work against "all the possible others" all the more reason that it has to work against the existing players.


Ours point of view (about AI improvement, testing) are really different, that's all i can get out of this discussion.

I cant argue more, i havent other arguments

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#13 » 01 Aug 2017, 07:31

Yes we can agree that we disagree.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
LuBeNo
Autonomous Entity
Autonomous Entity
Posts: 532

Re: ELO corruption

Post#14 » 01 Aug 2017, 09:55

IMO there is a problem. It is not a really a big one, but it is there. (loosing / winning against the same person distorts the elo highscore - perhaps 5 or 10 places difference between the real elo and the highscore)

A problem I wanted to solve here: viewtopic.php?f=2&t=800
I still like the idea.
Image
My algorithm of life: if(self.tired) sleep(); else if(self.hungry) eat(); else follow(Jesus);

TheKidPunisher
Autonomous Entity
Autonomous Entity
Posts: 563

Re: ELO corruption

Post#15 » 01 Aug 2017, 14:47

pier4r wrote:It is like I say to you: go an write down all the possible sums of integer numbers that results in 100'000, by hand.

As in 99'999 + 1 ; 99'998 + 1 + 1 ; 50'000 + 50'000 , etc.

Good luck with that.


After re-read your post (and re-estimate time it could take), 100 000 is quite enough to «not even scratch all possibilities»

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#16 » 01 Aug 2017, 16:33

I just took a seemingly small number, but once you consider all the possible partitions of it, well, it is not small anymore.

https://en.wikipedia.org/wiki/Partition ... _theory%29
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

User avatar
0x1de
Automaton
Automaton
Posts: 156

Re: ELO corruption

Post#17 » 02 Aug 2017, 02:19

We deploy 10 games at a time because we want some action and feedback on how affective our changes have been.

If there are duplicates of a certain map, that's fine if the opponents are different because it helps prevent specialisation against individual players.

On the other hand, if it's the same opponent on the same map with the same AI, it's just doubling (or tripling) the elo gained or lost.

I find that, in master league, I change an AI after a match and then have several games pending using the old AI.

As we say in English, "c'est la vie".

Captain Spock (Castlevania)
Automaton
Automaton
Posts: 177

Re: ELO corruption

Post#18 » 02 Aug 2017, 09:05

There is a problem to fix. One should not meet another player multiple times on the same map in one pending queue. It is unfair because the if the first player lose, he gains no additional information but lose more points.

pier4r
Skynet
Skynet
Posts: 3381

Re: ELO corruption

Post#19 » 02 Aug 2017, 10:27

For reference: viewtopic.php?f=8&t=140&p=8336

I almost finished the analysis (I stopped in May, one day I have to finish it and make an article). I can tell you though that if the game would give out permutations of the sequence of the maps instead of random maps, the average time a player will play the same opponent and the same map will go down.

In short: I do agree that, in one queue run, being matched against the same player on the same map brings no additional info. For this for example I adapted my queue deployment strategy: I deploy until I get a map twice.

To adapt this generally one quick idea (because if I understood GFX the code that selects the maps cannot quickly check the past info of the players) is , instead of doing:

Code: Select all

next_map = random_map_id();


the following

Code: Select all


pick_next_map_from_permutation() {
  if ( map_permutation_element_next_exists() ) {
    return map_permutation_element_next();
  }
  else {
    maps_produce_permutation();
    return map_permutation_element_next();
  }
}

maps_produce_permutation(){
  list map_permutation = shuffle_list( (syp, sw, cod, mp, mind, km, split, barred) )
}



In this way the probability for a player queueing 10 games consecutively (otherwise it is not true) to get the same map twice/thrice is quite lower than the actual one.

This is valid with a low active playerbase. If there are 300 persons queueing at the same minute, it will be essentially random.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

TalonZa
Script
Script
Posts: 24

Re: ELO corruption

Post#20 » 02 Aug 2017, 18:08

pier4r wrote:For reference: viewtopic.php?f=8&t=140&p=8336

I almost finished the analysis (I stopped in May, one day I have to finish it and make an article). I can tell you though that if the game would give out permutations of the sequence of the maps instead of random maps, the average time a player will play the same opponent and the same map will go down.

In short: I do agree that, in one queue run, being matched against the same player on the same map brings no additional info. For this for example I adapted my queue deployment strategy: I deploy until I get a map twice.

To adapt this generally one quick idea (because if I understood GFX the code that selects the maps cannot quickly check the past info of the players) is , instead of doing:

Code: Select all

next_map = random_map_id();


the following

Code: Select all


pick_next_map_from_permutation() {
  if ( map_permutation_element_next_exists() ) {
    return map_permutation_element_next();
  }
  else {
    maps_produce_permutation();
    return map_permutation_element_next();
  }
}

maps_produce_permutation(){
  list map_permutation = shuffle_list( (syp, sw, cod, mp, mind, km, split, barred) )
}



In this way the probability for a player queueing 10 games consecutively (otherwise it is not true) to get the same map twice/thrice is quite lower than the actual one.

This is valid with a low active playerbase. If there are 300 persons queueing at the same minute, it will be essentially random.


Pier, I agree you can "work around" the problem. But why should I? The point of this request forum is to request functionality that will make the game play experience better.

Id rather be able to safely deploy 10 matches and get back to life. Deploying only 2 or 3 until I get tje same match is just cpunter productive.

In the mean time, while we are waiting for GFX to solve the issue, I will ise your solution.

Return to “Feature Requests”

Who is online

Users browsing this forum: No registered users and 1 guest