Platinium Rift - 1v1 Results on the General Leaderboard

AI Battles
#1

Hello everyone,

Like most of us, I have a lot of fun playing to Platinium Rift Challenge. As the other multiplayers challenges before, it’s very fun and very addictive.

I played a lot of RTS when I was in highschool and it’s really nice to find again that fun here.

However, since the beginning, I’m a little surprise to see that I even if I beat a lot of top 15 players (It was before Jeff06 was here :)) in training in 1v1 (> 80% win on 10 games), my rank was not bad but not great.

So, as I saw on the chat some people made some stats about players fight, I decided to do it too, to see what was wrong with my results.

I got this:

*************
Resultats 2 joueurs
1 -  51/64(79.69%)
2 - 13/64(20.31%)
*************
Resultats 3 joueurs
1 - 55/112(49.11%)
2 - 26/112(23.21%)
3 - 31/112(27.68%)
*************
Resultats 4 joueurs
40/125(32%)
18/125(14.4%)
36/125(28.8%)
31/125(24.8%)

So, “OK, I have to improve my results with 4 players”. Now I wanted to compare thoses results with top 10 players. And here’s what I found on 3 of them:

Resultats 2 joueurs
1 - 28/96(29.17%)
2 - 68/96(70.83%)
*************
Resultats 3 joueurs
1 - 46/152(30.26%)
2 - 70/152(46.05%)
3 - 36/152(23.68%)
*************
Resultats 4 joueurs
1 - 50/180(27.78%)
2 - 66/180(36.67%)
3 - 50/180(27.78%)
4 - 14/180(7.78%)


*************
Resultats 2 joueurs
1 - 32/118(27.12%)
2 - 86/118(72.88%)
*************
Resultats 3 joueurs
1 - 41/144(28.47%)
2 - 66/144(45.83%)
3 - 37/144(25.69%)
*************
Resultats 4 joueurs
1 - 33/157(21.02%)
2 - 86/157(54.78%)
3 - 29/157(18.47%)
4 - 9/157(5.73%)


*************
Resultats 2 joueurs
1 - 11/50(22%)
2 - 39/50(78%)
*************
Resultats 3 joueurs
1 - 28/63(44.44%)
2 - 23/63(36.51%)
3 - 12/63(19.05%)
*************
Resultats 4 joueurs
1 - 27/75(36%)
2 - 30/75(40%)
3 - 17/75(22.67%)
4 - 1/75(1.33%)

So I find out what I noticed in training mode: They’re not good on 1v1,
but they’re quite good to finish 2nd in 1v1v1 and 1v1v1v1.

I know now (and I share that information with you:)) that if you want to grab some places in the ranking, you have to improve your games with 3 and 4 players in order to save a good place and not finish last. This means most of the time in those games, you don’t want to try to win, you just save what you can.
So Ok it’s a challenge to code this logic, and I will try to do my best to achieve it.

That being said, I think in RTS, 1v1 should be considerated as the “king” battle mode. If you want to know who’s best between 2 players at starcraft, they play against each others 3 or 5 games and you have your answer.
It’s the only mode which is fair and where there’s no luck: If you both spawn 1 POD in the same place, both will lose 1 POD. No third party will profit from that. If you don’t try to win, you just lose.

Here, the results in 1v1 seems to be irrelevant to the ranking, and I think it’s a bit sad: You can be ranked #300 even if you have more than 70% wins in 1v1 against most of people with better ranks. I think it should be results in 1v1 which should matter the most in the rankings.

It’s just my opinion.
What do you think about this?

3 Likes
#2

be carefull, thoses stats are only a percentage of win. it doesnt’t give you informations about “enemies” rank.

so you can win 90% of your match vs the 300th player, you will win less points than 20% of victory vs first player… :wink:

#3

Yes of course! I’m not saying that if you have 90% win on 1v1 at 300th, you’re better than the guy who has 30% win on 1v1 at 5th.

My point is, for the case you explain (90% wins at 300th), you should be ranked higher than 300th, let’s say 150th, with players against whom you will have 45-55% wins on 1v1. Because I think the only fact you’ve 90% win at 1v1 proves that your AI is better that the ones around.

On the contrary, if you have less than 25% of wins in 1v1 against people of your rank, I believe it proves your rank is too high. Even if you manage to save 2nd or 3rd place when you play 1v1v1 or 1v1v1v1 against those same people.

If there are 3 differents rankings for every kind of games (1v1, 1v1v1 and 1v1v1v1), and if you merge these ranking into a global one, in my opinion, for RTS, your ranking on the 1v1 games should have more weight than the others. Because like I said, it’s the more fair and “free-luck” mode. While now it seems (but maybe I’m wrong) it’s quite the opposite.
Maybe the first reason is simply there are less matches played in 1v1 than in 1v1v1 or in 1v1v1v1.

#4

The ranking seems based on the True skill system. So if you want to know how it weight in each match, better dig into that.

But I think you’re right, for me too 1 VS 1 is more fair because on 3 or 4 players, bad luck happens much mre often

1 Like
#5

I’m perfectly agree with that, it’s not really relevant to mix 2, 3 and 4 players.

We should have at least a two leaderboard :

  • 1vs1
  • multiplayer : 3 players and 4 players (best would be to have one leaderboard for each).
#6

The challenge is to be good in any situation …

#7

Of course, the challenge is good. As a software developer for years, it’s refreshing to try coding an AI.
I really like coding game for that : they providing an amazing plaform when you like coding.

It’s just some ideas to improve the system, and make anyone happy with the leaderboard.
Like this with a 1vs1 ranking there is no dicussion it’s fair. £ math with every contestant and there is no possibility to argue.
Morever you can run all the match is a fair amount of time.

Multiplayer is still a bit random, unless you can run all the possibility but it’s qui long.

#8

The problem is that once you have played your 100 first matchs you have no chance to fight 1v1 with your neighbors in the leaderboard. The only 1v1 matchs you play is vs people who just pushed their AI and their ranking is not relevant yet.
So if you are good in 1v1 you always get up on the leaderbord in first place, and then slowly decrease…

2 Likes
#9

Yeah that’s what happen with me.

If there were more leaderborad, that could be more fair. And more interesting to see how your program behave in each situation.