Analyses of the ratings - Spotting the issues

Age of Empires II: DE II - Discussion
352

Man you are new here it seems, they implemented a system that was wrong since day 1, it had good things on it like the boost to increase rank faster in case of smurfing to prevent noob bashing, but unlike voobly it wasn’t well implemented and it caused a HUGE inflation in both ladders cause the higher rank reached on whatever ladder was used as reference to start, we used to have 2.2k players with 10 loses 1 victory in the 1x1 ladder cause they were 3k tg elo.

Also they never thought on team games, the elo system for team games usually gives less points for TEAM GAMES cause there are several factors that determine the victory rather than skills, so since GAMING ZONE(1998) for a 4x4 you would get 4 points per victory, on voobly at 2014 you would get 8 points, the ladders there were controlled cause even stacking hard, the number of games needed to reach higher ranks were too much, BUT here on DE since day one for a 4x4 win you were getting +17 points> this by itself created the 3k elo players in a matter of 3-5 months.

Now they removed the boost, so the smurfs can noob bashing forever cause they need to play hundreds of games to reach their true rank and at the same time they can point trading higher players to get more points, this was the biggest flaw on their latest change, the only positive measure was reducing the number of points per victory on equal games around 11 instead of +17.

There is another thing that it seems that only me is pointing out, the 1x1 ladder has a ceiling limiting the higher ranks to increase more, starting from 2200 the players starts winning and losing less points and it increases the higher they are, a match btw 2 2500 players results in a 6 points exchange and a 2600 players only get 4 points vs a 2580 lol, that ceiling doesn’t belong to the 1x1 ladder it should have been implemented on the TG ladder(this ladder has no ceiling or limiter that is why you see 5k elo players), as result the 1x1 TOP ladder isn’t accurate as you may see 20 players close to each other while the truth is completely different.

353

I’m aware of that, I was one of the people suggesting ways it could be improved. But some people insisted a straight average of each team’s ratings would be fine, and that was what they went with when they changed it, causing a new set of problems.

1 Like
354

They might, but in my experience it’s not certain, development teams can be very different depending on the preferences of the hiring manager for the types of people and backgrounds they like to recruit from.

1 Like
355

I feel the whole purpose of this thread is to obfuscate the discussion about the matchmaking and rating systems.
Because @WoodsierCorn696 sets references to it all the time, and the title says it’s about ratings only, and it’s not readable.

It’s not readable, because extensive topics like rating system, matchmaking, player misbehaving are just scattered in all posts.

The initial post is not updated either.

I highly recommend closing the thread.

1 Like
356

I just started to make some tests with the new handicap system to find the correct value for ∆ in @Mercy9545 s proposal here .
My Idea is, if we can find good elo difference approximations for the different handicaps, we might have a chance to find the ∆ value without the need for human tests (and the flaws that come with them).

The concept is quite easy: I let a single DE Estreme AI fight against a team of 2 or more Extreme AIs and increase the Handicap until it’s basically “even”. I repeat this several times to get confidence.

I chose to give all civs Magyars, as this is the civ that can play all TG positions and I didn’t wanted to interfere with my experience by giving one side an advantage.

My Results so far:

Single AI vs 2 AIs:

5 Games with 125 % Handicap: 1 Won, 4 Lost
5 Games with 130 % Handicap: 4 Won, 1 Lost

Single AI vs 3 AIs

8 Games with 160 % Handicap: 4 Won, 4 Lost

Single AI vs 4 AIs

7 Games with 195 % Handicap: 5 Won, 2 Lost
7 Games with 190 % Handicap: 1 Won, 6 Lost

Single AI vs 2 AIs 100 % and 1 130 % AI

3 Games with 195 % Handicap: 3 Won, 0 Lost
4 Games with 190 % Handicap: 3 Won, 1 Lost

Unfortunately it looks like the Handicap system is quite linear with the number of opponents. It would be easier if it was linear with the logarithm of the opponent numbers. But this shouldn’t stop us from getting usable results. We only need a way to get the elo difference between an 100 % ai and 125-130 % ai and a 160 % ai. If we get these we can calculate the ∆ value.
The way how this then works is that we will calculate for each different 2-set of (1v1, 1v2, 1v3, 1v4) a a ∆ value and compare them. If they are quite close to each other we have a good indicator for our method being reproducable and therefore can use it for team elo.

One interesting fact about that method is that we actually can calculate a “Team elo” then, not an “average Team elo” what means that it would be possible (in theory) to make TGs with 3 v 4 or so that are balanced.

Somebody has an Idea how we get the Elo for 130 % AI?

2 Likes
357

Play it against people who play the meta on any given level. I think that last I checked, the 100% extreme AI was evaluated at about 1.2k, so start there, and more up in 100 elo blocks, or even 50.

358

You really overestimate the extreme AI. My own guess for the extreme AI is about 900 elo.

359

How recently have you played it on arabia? It’s definitely not a 900, it plays significantly better than your average 900 would.

360

I am not really into playing the AI, but about 2 year ago, when i went back to this game, into DE, i was pretty terrible on the ladder, but i was able to beat the extreme AI in 1v2 on normal settings. The AI did get some improvements, but no real big ones (afaik). But also the player base improved. A 900ish elo player 2 years ago has no chance of winning against a current 900 ish player. For that reason T90 also see LEL not as <1000 anymore, but he wanted to lower the upper limit, because the players get better.

The strategy against the AI is just hit them early and they will never recover. Aggressive 800-ish players might even able to beat the extreme AI, while more passive, boomy players at 1000 elo might struggle against the AI. But on average it is about 900-ish when you start beating the extreme AI regurally.

1 Like
361

i’m 1200 and lose to it from time to time. but yeah with certain strategies ai is easily beatable.

362

It has improved a lot in 2 years. Also, they get better if you don’t rush them or anything, because they boom really effectively. Their potential for a good game is better than 900 I think.

363

i would find it interesting if they put a bunch of AIs on the ladder to see where they end up. ideally with some names which dont make them immediately recognizable. probably not any with handicap though, because that feels like cheating

3 Likes
364

They could possibly add this to the next PUP. Adding bots with various handicaps there. (It’s important that the players don’t know wether they face an ai initially)
As nobody is forced to play PUP if it is communicated right I see only benefits of this.

365

Barbarian is considered the best AI (better then the ingame AI) and Viper stated Barbarian is about 1200-ish. This back ups in game AI is at least less then 1200.

I would be fine to see this on the normal ladder as well. No need to be PUP only. Having some bench mark on how well the AI is performing is fine to me. I dont think many players will really see the difference while playing, except in the communication. Pro players would probably spot it instantly, but in the range of 800-1200 no one will really spot it.

I agree on the 'use unrecognizable names part.

The endgame of the AI is pretty strong with the macro of all the units for people at this level. Hit them hard and hit them early. But these are also the reasons why lower rated people tend to fail against the AI.

2 Likes
366

I’m not sure how people would feel about losing elo to an AI if they ever found out.

367

you could make it so that the human players dont lose/gain elo when they play against AI…might lead to an issue where people get annoyed that they wasted time though

I think an issue with having AIs with handicap on the ladder is that some players like to watch their recorded games and compare build orders with their opponents. this wont work if it’s an AI with handicap

one option would be to have options
-i accept being used as a benchmark for AI
-i accept being used as a benchmark for AIs with handicap

368

Viper doesn’t know what 1200 means. He’s much too high ranked to evaluate that.

2 Likes
369

That’s true, and that barbarian is the best AI is also already outdated.
Viper don’t know nothing except his ever same game style.

370

Just a question, what about a handicap torunament? The less handicap you have, the more points you get from a victory.
So there could be a big group of like 10 pros fighting in a league and deciding individually what handicap they play (without knowing the opponent handicap before hand). For each victory you get 1 point + 1 point for each 5% handicap less than 200 %.
It would be interesting to see how pros would chose their handicaps against like players with 2000 elo or so.

(the handicap system would make games naturally much faster so we would most likely see a lot of action aswell)

1 Like
371

To be clear: It isnt my opinion, because Viper said something. I said something. Afterwards i heard this from Viper.

I could also say cite T90, where LEL first was below 1000, but the level of players increases. What was 1000 in the past, is now 800. Based on that and my own experience when i was around that Elo, i am pretty certain about my opinion about which Elo should beat the 1v1 extreme AI.