matchmaking small improvement for psychology.

mumpsimus
Algorithm
Algorithm
Posts: 63

Re: matchmaking small improvement for psychology.

Post#21 » 02 Nov 2018, 18:03

I too initially thought that these ghosts would rise to a ranking reflecting their actual strength much more quickly, but the fact is that there are just *way* to many of them for that to happen. We are still seeing many overpowered ghosts from just after the first reset at low Elo rankings, let alone the second one.

pier4r
Skynet
Skynet
Posts: 3381

Re: matchmaking small improvement for psychology.

Post#22 » 02 Nov 2018, 19:00

mumpsimus wrote:I too initially thought that these ghosts would rise to a ranking reflecting their actual strength much more quickly, but the fact is that there are just *way* to many of them for that to happen. We are still seeing many overpowered ghosts from just after the first reset at low Elo rankings, let alone the second one.


This is a good point. For example if there are few active players, they are against many ghosts that are relatively strong and ghosts that are not strong gets farmed and disabled. It makes the top league over time harder. It is not bad, it is hard and not psychologically ok.

So once again I suggest the point of disable (better: do not consider) ghosts of players with a score over 400 points over the score of the ghost.

In code

Code: Select all

if ghost.score + 400 < ghost.player.score then
  do not consider the ghost as a valid one
end


In this way only when a strong player drops in rating (or is inactive) all the lower ghosts gets reactivated. This also is a relatively nice solution for the related thread linked before. viewtopic.php?f=2&t=1433

edit: also about the prediction of GM becoming harder (viewtopic.php?f=7&t=1354&p=17222&hilit=reset#p16835) (although it took 2 seasons and a threshold adjustment plus matchmaking adjustment)
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

Bockwurst
Script
Script
Posts: 25

Re: matchmaking small improvement for psychology.

Post#23 » 02 Nov 2018, 20:55

Getting rid of resets is a valid idea. I wouldn't miss them.

IF we decide to keep resets then there's two options to sanitize ghosts:
1. Remove ghosts that have reached some criteria
2. Only allow the creation of ghosts if some criteria are met

Personally I prefer contraceptives over abortion ;)

Possible ghost creation criteria:
- winrate below 90%
- ghost submitted in the past day/hours less than X

pier4r
Skynet
Skynet
Posts: 3381

Re: matchmaking small improvement for psychology.

Post#24 » 02 Nov 2018, 21:40

We already discussed the win rate idea and it would keep away ghosts of very particular players like yasgel.

Therefore I mixed the bockwurst idea with the score "compared to the actual score player".

Anyway mixing more ideas. There are several players that over time (since February 2018) asked that the ghosts were the closest to the last deployed ghost.

So here another idea that would satisfy many others as well.


Keep only trailing ghost active . Say the last 50 deployments (the number can change) of the player are active. If one of those gets disabled (loses points) then the 51st deployment is activated, if another gets disabled then the 52nd deployment is activated and so on.

In a database this would mean: keeping a table with ghosts id and a boolean if they are active or not. At each new deployment from a player a routine goes disabling the now-to-be 51st last active ghost.

And now let's check the points.

The active ghosts are closer to the last deployed ai. True. Although if a player doesn't play and his ghosts gets gradually farmed means that ghosts distant from the last deployment gets activated. See good bot at the moment.

All the ghosts will be active sooner or later . This is true unless the player keeps playing or some 50 ghosts are never farmed.

If the player is strong and climbing fast the lower rated ghost (underranked) are not going to farm everyone else. True, as only the most recent deployments will be active and if the player reached a temporary plateau more or less all the ghosts are close to his actual score .
Drawback. If the player drops like a stone for some tests while spamming matches his high rated ghosts are also disabled .



If this is complicated I also like the last idea of bockwurst, rather than the winrate. I use his idea with a small change.

The possible ghosts are deployed like now. No deployment is lost (this for possible moments of low activity of the playerbase in a certain score range).
Anyway of those deployments only one ghost per day (or hour) is active .

So say that the player gfx47 played a lot and in 3 days he swinged from 1400 to 1800 and back.

The first day was 1400-1600. The second 1600-1800 the third 1800-1400.

Only one randomly picked ghost from day1 , day2 and day3 are active . (Of course If they are eligible).
As soon as one of those three ghosts gets deactivated due to being farmed , another ghost in that day (if eligible) gets activated.

If all the ghosts in a day are farmed and disabled , that's it. No other ghost is going to be activated.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

mumpsimus
Algorithm
Algorithm
Posts: 63

Re: matchmaking small improvement for psychology.

Post#25 » 02 Nov 2018, 21:57

I like Bockwurst's idea of only creating a new ghost for a player at most every N hours. This will mean ghosts are more likely to have differences, and will not create many more ghosts for folks who spam a lot of games with an identical AI than for those who deploy fewer games.

I think this, in combination with getting rid of resets would improve things a lot.

pier4r
Skynet
Skynet
Posts: 3381

Re: matchmaking small improvement for psychology.

Post#26 » 13 Nov 2018, 00:14

I'd say the recent move of gfx keeping only ghosts if the md5 if the ai differs should prune a lot of similar (underrated) ghosts. As far as I understood, he picked the highest score ghost for each md5 generated.

For the moment we can see what happens given this change.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

sollniss
Automaton
Automaton
Posts: 178

Re: matchmaking small improvement for psychology.

Post#27 » 18 Nov 2018, 15:23

What's the argument against Glicko/Glicko 2 again?

Why not use something mathematical stable instead of some weird homebrew?

pier4r
Skynet
Skynet
Posts: 3381

Re: matchmaking small improvement for psychology.

Post#28 » 18 Nov 2018, 22:46

Holy ignorance. "weird homebrew" lol.
http://www.reddit.com/r/Gladiabots/wiki/players/pier4r_nvidia_shield_k1 -> Gladiabots CHAT, stats, insights and more ;

sollniss
Automaton
Automaton
Posts: 178

Re: matchmaking small improvement for psychology.

Post#29 » 19 Nov 2018, 13:48

Can you answer my question?

Return to “Feature Requests”

Who is online

Users browsing this forum: No registered users and 2 guests