Alternative to game quality indicator - Handicaps/Bonuses

@boom said in Alternative to game quality indicator - Handicaps/Bonuses:

  1. 500r dude gets x1.5 mass
  2. 500r dude continually feeds mass to 1500 teammate
  3. ?????
  4. profit

Also no idea why you think giving a low rated guy more resources and buildspeed would increase their effective skill, if anything it would turn them into a mass donation machine akin to cheating ai

Although your numbers are off, I understand your concern. It may be necessary for a handicap system that units and/or resources cannot be shared between team mates. This is not out of the question.

Your axiom is that a FAF game must always be balanced so that everyone has a 50% chance of winning rather than being meritocratic and rewarding improvement in skill. It doesn't matter how good I am, I will always have the exact same probability of succeeding.

You basically want to turn the game into literal rock-paper-scissors. And that's giving you the benefit of the doubt and assuming you have actually found the perfect values between ratings to reach your intended goal. This is not a replacement for trueskill, it is literally anathema to trueskill. This is a mod for the game, not a new rating system.

@ftxcommando said in Alternative to game quality indicator - Handicaps/Bonuses:

In your scenario the 500 is overrated as he is functionally a 0 or even a negative as far as trueskill is concerned (but it cant make these observations without reviewing games like this). If you were able to defeat 2 1000s (who we assume are certainly 1000s) independent of your teammate, then you would at minimum be a 2k player (and underrated at your present 1500).

I don't think your assumption here is correct. I had a game earlier today with me (1500) and a 600 player against two 1200s who double-teamed me. My 600 teammate did basically nothing while I was double teamed. I'm not certain, but I suspect I would have faired better if I simply spawned by myself. If the game started as the same 2v2 but I was in command of both ACUs (therefore more like a 2v2), I am fairly certain I would have won, or at least matched the other team (therefore been close to 1200x2 meaning 2400).

It's important to note that truskill was developed for Halo, a first person shooter with equal numbers of players on each team (say, a 2v2 or 4v4). Truskill is certainly useful for FAF, but FAF is not an FPS and there are significant differences. (On a 5x5 map, could a 2000 player with 1 ACU be expected to beat 20 players of 100 rating with 20 ACUs? I think not.)

@ftxcommando said in Alternative to game quality indicator - Handicaps/Bonuses:

Your axiom is that a FAF game must always be balanced so that everyone has a 50% chance of winning rather than being meritocratic and rewarding improvement in skill. It doesn't matter how good I am, I will always have the exact same probability of succeeding.

You basically want to turn the game into literal rock-paper-scissors. And that's giving you the benefit of the doubt and assuming you have actually found the perfect values between ratings to reach your intended goal. This is not a replacement for trueskill, it is literally anathema to trueskill. This is a mod for the game, not a new rating system.

I disagree. I would prefer doing away with game quality indicator altogether. I - a high skill player - will be at an advantage if no game qualities are ever calculated, and if player global ratings are hidden (so players cant estimate game quality themselves).

This "balance" axiom is YOUR doing. I would rather the balance simply disappear.

No, my assumption is correct. Your assumption that the result of a single game is evidence of the long term failure of trueskill is what is flawed.

Your analogy about trueskill is also flawed because it can only make a relativistic distribution based on the data it has. If you have a distribution based on 5x5 maps, then it is accurate for gauging that. If you suddenly add 20x20 maps, then you have added an error factor for the rating. Since tmm eliminates the ability to select for slots and maps (but it keeps the ability to select for teammates), it is about as close as you can get to rating the quality of that individual as a teammate.

So: since there is no situation where trueskill needs to account for data of 1 2k player playing 20 100 rating players on a 5x5 map, it is irrelevant to the situation. All that matters is 2v2 capability on the curated pool of the matchmaker.

So can a 2k player beat 20 100 rated players on the maps in tmm? If he can't, then he isn't a 2k according to the construction of this trueskill environment. If he can, then he is. Luckily this is not a possibility in matchmaker so it isn't a consideration for the system.

@ftxcommando said in Alternative to game quality indicator - Handicaps/Bonuses:

No, my assumption is correct. Your assumption that the result of a single game is evidence of the long term failure of trueskill is what is flawed.

Your analogy about trueskill is also flawed because it can only make a relativistic distribution based on the data it has. If you have a distribution based on 5x5 maps, then it is accurate for gauging that. If you suddenly add 20x20 maps, then you have added an error factor for the rating. Since tmm eliminates the ability to select for slots and maps (but it keeps the ability to select for teammates), it is about as close as you can get to rating the quality of that individual as a teammate.

So: since there is no situation where trueskill needs to account for data of 1 2k player playing 20 100 rating players on a 5x5 map, it is irrelevant to the situation. All that matters is 2v2 capability on the curated pool of the matchmaker.

I disagree. I think truskill is quite good. My OP only questions the validity of the game quality indicator. The game quality indicator is NOT truskill. We shouldn't pretend that it has the same reliability. Remember, my OP suggests handicap based on truskill. Nobody here - not me, not you - is trying to say truskill doesnt work.

I thought the game quality indicator was literally the fed likelihood of a game result based on given mu and sigma?

@ftxcommando said in Alternative to game quality indicator - Handicaps/Bonuses:

I thought the game quality indicator was literally the fed likelihood of a game result based on given mu and sigma?

In the context of equal teams in FPS games, yes. However, truskill was never intended for use in RTS games or in game with unequal numbers of players in each team.

The truth is, the game quality indicator is nearly arbitrary. It basically just averages the truskill of each player. IT doesnt account for anything else, despite there being a lot more to account for in RTS games.

Can't the TrueSkill algorithm be made to work faster?

If there is a significant rating difference between team members, let it learn multiple times from the same game for the lower rated player.

Example 500 & 1500 vs 2 x 1000. 1500 player gets one normal rating adjustment. 500 player gets rating adjustment (1500/500=) 3 times in a row or multiplied 3 times. Now a 500 player who should be 800 rated gets there faster and less people post on the forums, otherwise his rating drops faster so next time they play 2x 900.

ya that's why trueskill has deviation to learn rating faster

@funkoff said in Alternative to game quality indicator - Handicaps/Bonuses:

In the context of equal teams in FPS games, yes. However, truskill was never intended for use in RTS games or in game with unequal numbers of players in each team.

Trueskills is a mathematical model of player skill. It does not matter in which genre. What can matter however, is how a certain game mode works, so yes it could have problems with how Trueskil expects player skill to add up in teams.

Your suggestion however, does nohing to fix that problem. I turns gameplan from a good player trying to carry a bad team to a good player trying to exploit the bonuses of his team mate in the best way possible.
It makes the entire game unpredictable and inconsistent (you now longer know much damage ANYTHING does to each other) and in a way unlearnable, or rather increases the skill gap massively because you now need to learn how every units properties scale with malus/bonus level.
And it adds an entire dimension of balance maintenance work. Not gonna happen.

So, here are the two solutions to the problem:

  • Change Trueskill to model how bad players can be a net negative for the team by preventing a better player from gaining more mex spots. Not gonna happen.
  • Change the game so that a dying low rated player no longer gives any additional mex spots to the team. For example by removing core mex spots on death. However, that changes a whole bunch of other dynamics related to player deaths and share conditions.

The biggest problem with putting a "thumb on the scale" to benefit one player over another is that it interferes with both players' natural development of an understanding of eco balance. The stronger player is going to have to fight against his own brain to make sense of a sluggish balance and the weaker player is going to be denied the opportunity to develop that aspect of their game sense. Which means neither player is going to get better. It will be not only a frustrating experience but also a pointless one.

It would make more sense to have a handicap in the form of one player starting with prebuilt structures. From there, the game would be completely normal: every unit and building would behave exactly as normal.

It's true that the stronger player could catch up to the weaker player and then the game would basically be over--but that just means the weaker player needs to learn how to play from an advantage. This wouldn't be for TMM of course, it would just be for custom matchups. It wouldn't just be about balancing 1v1s or 2v2s. You could balance a 2v1 this way, or you could balance a 6v6 if the weakest players got to start with an extra factory and some pgens. You would just have to expect them to behave as gentlemen and not give their factories to a higher-rated teammate.

You might be able to modify the "prebuilt bases" code to look at player ratings in order to decide how much each person gets of the prebuilt stuff. So if you this SIM mod running (let's call it "Balanced Bases") and prebuilt buildings are off, nothing happens. But if prebuilt buildings are on, the stronger players get less and the weaker players get more.

Any idea that changes the probabilities of a games outcome significantly based on player rating, be it by unit balance, or by adding prebuilt structures, kills Trueskill ability to work correctly.

Edit:
Let me clearer on that:

The rating of a player must never influence the game rules, because otherwise you break the maths and the world explodes.

And this is probably true for most rating systems.

This post is deleted!

@humanpotatoe said in Alternative to game quality indicator - Handicaps/Bonuses:

trueskill is useless for anything other than 1v1, and even there whoever controls the map pool and decides what the reclaim values are on each map effect that rating. whoever controls it now loves rocks and spam heavy gameplay

You are trolling right? Please leave this discussion if you have nothing of value to add.

This post is deleted!

On a COMPLETELY related note, why is there no downvote option on the forums?

So that we can reinforce positive mindset and properly show how caring of community we are.

i love this idea it has my vote to buff lower rank players with unimaginable bonus's