forum

Team Match

posted
Total Posts
56
This is a feature request. Feature requests can be voted up by supporters.
Current Priority: +1,061
show more
-Chizuru-
Seems to be cool. ppypls
Ous
I'm agree with this idea!
Ashina Izu
Nice idea o/
Noxise
Count me in ! o/
abraker

Arrcival wrote:

TP = (1000+sqrt( Team score 2 (winning team) - Team score 2 (losing team))/100
Evaluating proposed formula. Check your parenthesis.
Topic Starter
Arrcival
Parenthesis missing bcz of the (**** team), thanks^^
Pachiru

abraker wrote:

Arrcival wrote:

TP = (1000+sqrt( Team score 2 (winning team) - Team score 2 (losing team))/100
Evaluating proposed formula. Check your parenthesis.
Oh le type, il t'a pas respecté tho



Sinon, je valide mon accord mdrlol <3
Antideveloppeur
Please ppy !
Nelyus
I agree with that idea it's awesome !!!
;)
snz
I Agree too but you do not think you have something similar?

And it is also almost the same as the Multiplayer.
And yet.
I dunno if Peppy add something as like another thing he already has.
So, It can be something just for supporters or something like this.
Topic Starter
Arrcival

Ppus wrote:

I Agree too but you do not think you have something similar?

And it is also almost the same as the Multiplayer.
And yet.
I dunno if Peppy add something as like another thing he already has.
So, It can be something just for supporters or something like this.
The design is similar, yes... Don't know future design of osu!next so...

Yes it's a multiplayer, but all the players dont have to connect simultaneous to play for his team..
dunno what Peppy need or want so I only propose a system quite different of the multiplayer actual system
abraker
Woah, that's a lot of stars. This is a nice idea, but the formulas need work.

Before I start, let me mention that I have a similar request which is asking for a team based scoring system. The only difference is that it allows teams of any level to be matched. The formulas in that request need some work and is likely not to receive the same attention as this request. If you decide to allow teams of any level to be matched, then I can let the mods kill that request. I will have to figure out the formula that allows to compare plays by lower ranking player to higher ranking player on an equal level.

Now let's get down to the formulas:

  • Team level = (Average stars of the 100 best performance of player 1, added to player2, 3, n)/n
    This formula has 2 issues
  1. You are taking the top 100 best performances of some player. If you look at players' top performance section, you will notice that performance quickly degrades as you go down. The 100th performance would be anywhere between 7/8 to 1/2 of the top 1 performance. Moreover, what if a player has several 200pp performances and the rest 100pp performances and lower? Should that player drag the team lvl down because of it even if that player can do much higher difficulty than what you formula says? I suggest taking the top 20 or top 10 performances to provide better results of what the player is capable of.
  2. You are going by the stars, which is map difficulty for the top performances. Suppose a player has a top of 20pp D ranks on 5* maps. So the formula will justify that player can do 5* maps? You might as well get rekt.

    TP = ( 1000+sqrt( Team score 2 [winning team] - Team score 2 [losing team]) )/100
    This formula really is this:

  3. TP = 10 + sqrt(avg(winning team) - avg(losing team])/100
This has several problems:
This is SOOOO farmable. The loosing team can just be losing combo on purpose, giving any team they get matched with points. And they wouldn't even get penalized for it. Brilliant!
There is unfairness if the teams are imbalanced. Suppose it's 1 vs 4. The one player get 3M and the four players get 1M each. According to your formula, the team of four wins. Where is the fairness in that? The team of four players have an advantage in quantity. It's more impressive that the one player manages to get a score much higher than any of the team of four individually.
Now to solve the problems:


  • Team level

    This is the core of the formula you see. It equals to 1. That means what ever it is multiplied by would be equal to the value you multiply this by. The ratio you see in the formula is the ratio between the achieved score for the map divided by the max possible score achievable for the map, taking mod multipliers into account. According to this formula, a play of 750,000 out of 1,000,000 on a 5.2* map would be equivalent to a 3.55* play. The equivalent star diff goes up exponentially as the ratio approaches 1:1 until it reaches the difficulty of the map. You can play with the number here. This would be calculated for every play in the suggested top 20 or top 10 and then averaged to come up with the player's star level. Then this will be done for every player and that will be averaged together. So the formula would be this:


    Team points
    This one is more complicated. This can be done 2 ways depending on what you think is more fun to see. Would you rather see players getting more points the wider the score gap or more points the closer the score gap? If you think about it, the one with the closer score gap sounds counter-intuitive. How can a play of 1,000,000 to 999,999 be worth much more than a play of 1,000,000 to 100,000? With the current intuition, the team with 1M points should get much more points if the other team scored 100,000 than if the other team scored 999,999. But what would you rather see? An extremely close match which raises the stakes into a hype level as high as cookiezi's 700pp play, or just another looser vs winner match? In my opinion the points should be counted based on what kind of performance it gives to whoever is spectating, not for the teams.

    I'll give you formulas for both ways, then you decide what will be more fun to see: a 1,000 team points extremely close call or a 1,000 team points overpowered winner vs looser.

    More points for wide score gap:
    This formula doesn't care who is winner and who is looser. The looser and winner is determined by the formula just by plugging in the results. The winner will receive a positive score result, and the loser will have a negative score result. Why negative? To penalize for loosing, as stated as one of the problems. Yes, there will have to be teams with negative team points. They should try harder to get those back :D

    Also notice how it's a ratio of means. That means (no pun intended) that the players on the team will need to dependent on each other more. One player can't simply carry the team. One player with a score of 1M defeats four players all with scores of 999,999. This will make it fair for both sides and both sides shouldn't worry about being outnumbered. If a player fails, that would count as a 0 and the player should not be able to recover. If a player disconnects, then it will should also count as a 0 to prevent players disconnecting on purposes to boost the team's score if the disconnected player has a low score. Nothing can be done with accidental disconnects, not that anything has been done about it in the past.




    Due to the relation between the two formulas, it can be simplified to Team A = -Team B or Team B = -Team A. This formula is based on the perception of gap between 2 values. The closer the gap, the more insignificant it is and vice-versa. I used scores from OWC 2015 to see what results this will give under score v2, and I suggest score v1 to be used for this formula. You can see the results here. As for mania, I suggest you multiply by 100 instead of 10.


    More points for smaller score gap:
    The formula for this would be:


    Like last time, in this case Team A = -Team B. You can play around with the numbers here.

    How I think players will behave to this kind of system makes things interesting. I see a lot of players bailing as soon as they think they can't make to the low score gap or will loose a good amount of points because they are lagging behind by some 10 points or so. However, the ones that go all the way through at achieve the low score gap will result in a rare event. A score ratio of 1,000,000 : 999,999 produces a huge amount of team points (230k), but this is very unlikely using score v1. If the ratio is 1,000,000 : 999,000 then it produces about 231 team points. Those 999 points would be worth a lot if targeted correctly, and that would be very impressive to see.

    There are several minor problem with this formula. One is the event of a tie. That has to be a case to be handled separately. In in this event, I think it's fair to give both team an equal amount of team points equivalent to sqrt(score) or something similar to this.

    Another one is that it is easier to reach a smaller score gap on easier map. In that case I suggest to multiply by star_diff.

    Last but not least, players targeting the small score gap can make it easy by breaking combo a lot. It would be possible to get a lot of points for low scores in some edge cases. To fix this, I suggest multiplying by accuracy, where accuracy is a value from 0.0 to 1.0.

    To make the formula more sensitive to difficulty and accuracy, I squared those.

    The final formulas after fixing the problems are the following and you can play around with them here

Well that's all. I'll wait for response/criticism/etc.
Topic Starter
Arrcival
Team level = (Average stars of the 100 best performance of player 1, added to player2, 3, n)/n
This formula has 2 issues
You are taking the top 100 best performances of some player. If you look at players' top performance section, you will notice that performance quickly degrades as you go down. The 100th performance would be anywhere between 7/8 to 1/2 of the top 1 performance. Moreover, what if a player has several 200pp performances and the rest 100pp performances and lower? Should that player drag the team lvl down because of it even if that player can do much higher difficulty than what you formula says? I suggest taking the top 20 or top 10 performances to provide better results of what the player is capable of.
You are going by the stars, which is map difficulty for the top performances. Suppose a player has a top of 20pp D ranks on 5* maps. So the formula will justify that player can do 5* maps? You might as well get rekt.
Totally ok, u rekt me ;w;

TP = ( 1000+sqrt( Team score 2 [winning team] - Team score 2 [losing team]) )/100
This formula really is this:
TP = 10 + sqrt(avg(winning team) - avg(losing team])/100
This has several problems:
This is SOOOO farmable. The loosing team can just be losing combo on purpose, giving any team they get matched with points. And they wouldn't even get penalized for it. Brilliant!
There is unfairness if the teams are imbalanced. Suppose it's 1 vs 4. The one player get 3M and the four players get 1M each. According to your formula, the team of four wins. Where is the fairness in that? The team of four players have an advantage in quantity. It's more impressive that the one player manages to get a score much higher than any of the team of four individually.
Yes it's farmable :< But when I wrote this i didn't think it will be teams with guys breaking their combos
Suppose it's 1 vs 4 -> At first I was thinking about matchs with teams with the same number of players...
If 1v4 is possible, average score v2 is more worth it than score v1 yup
My formula rekted


After, I test all your formulas and I have like them all but :

Team level is average of LVL(x) so?

I don't know what to chose between a wide or small score gap
But I think the small score gap option is the best because if guys break their combos intentionnaly with a system of "more points for wide score gap" will again be farmable : 900000 for 100000 will give a lot of TP^^

I really like all your formulas, thanks!
Can I use your formulas so and replace in the first post?
abraker
Yes, sure. I wouldnt have posted them if I didnt want you to use them ^.^

I just noticed a small mistake with Lvl(x). x is supposed to be the ratio, and the function has to be Lvl(diff, ratio), and not just Lvl(ratio). I'll fix that when I get on the computer later on

Arrcival wrote:

Team level is average of LVL(x) so?
You take the average of Lvl(diff, ratio) for the top10 or 20 scores to get player's average level. Then you take all of the players' level averages to get the team's level average.
Topic Starter
Arrcival
But
More points for wide score gap
or
More points for smaller score gap?

If a team win 850000/845000, so he win a lot of TP ok.
If a team win because they were very good at this match 950000/400000, they should win many TP too...

So I propose :

Final TP = big gap + small gap ?

I have other questions or problems about, don't want to write something false (bcz i'm french xd) so u can send me a pm website/ingame
abraker

Arrcival wrote:

So I propose :
Final TP = big gap + small gap ?
Looks interesting. In this case, the winning team will receive the minimum points possible when the score ratio is 1.259. That minimum score can vary due to map difficulty as well as average accuracy achieved among players' plays. The team winning an SS on a 4 star map guarantees the team will get at least 32 team points, and a linear increase to a minimum 72 team points on a 6 star map. This might solve a minor problem where the points are insignificantly small in some cases. This might work.

I noticed another minor problem concerning the small gap. For lower star maps, the slope as too steep, and it becomes easy to land thousands of team points. The only way I can think of to fix this is to apply a horizontal shift towards the left, breaking the symmetry how Team A/Team B translates to negative amount of points for Team B/Team A. To fix this, a piece wise function will have to be made. So the result is this:

And the graph which you can play around here looks like:

Here is the fixed formula for player level:

Arrcival wrote:

I have other questions or problems about, don't want to write something false (bcz i'm french xd) so u can send me a pm website/ingame
Sure, send me a pm if you want. I'll try my best to answer it.
Pituophis
This would be really cool.
Alaxandar66
It's definitely something I would enjoy, even as a casual player.

I support this all the way! :D
gigaiDX
Really well thought out and developed. Hopefully teams in osu!next will follow along the same lines!
ReaVix
I agree
Scarlet Evans
I like it! I didn't had the time to analyze it properly and try to pick some holes, but I will try :)

For now, have my star :D
WubWoofBacon
already been posted, but this one seems more useful i guess.
t/29780
abraker

WubWoofBacon wrote:

already been posted, but this one seems more useful i guess.
t/29780
Dont confuse this request with that one. That request only applies making clans, while this request applies to multiplayer, independent of whether clans are a thing or not
WubWoofBacon

abraker wrote:

Dont confuse this request with that one. That request only applies making clans, while this request applies to multiplayer, independent of whether clans are a thing or not
well whatever it is, it's a good idea(not that i read it all xD) so +4
Minsoo
I hope you know



e^ln2 is the same thing as multiplying by 2.

Can you just put in diff * (2e^ratio - 1).


Edit: Not gonna lie, this formula is terrible. The math behind calculating fair scoring needs to be revised.
Topic Starter
Arrcival

Minsoo wrote:

I hope you know



e^ln2 is the same thing as multiplying by 2.

Can you just put in diff * (2e^ratio - 1).
Yeah yeah, i didn't make the formulas but i know xD

Minsoo wrote:

Edit: Not gonna lie, this formula is terrible. The math behind calculating fair scoring needs to be revised.
I listen you
abraker

Minsoo wrote:

Edit: Not gonna lie, this formula is terrible. The math behind calculating fair scoring needs to be revised.
Can you explain the key points of what makes this bad? I provided 2 versions of the formula in the earlier posts and OP decided to mend them together. What kind of behavior are you expecting due to the results of a team match?

The reasoning behind the formulas can be found in the earlier post: p/4921364
Nyantares
Great idea! Would be really happy to see it in future updates :3
Minsoo

abraker wrote:

Minsoo wrote:

Edit: Not gonna lie, this formula is terrible. The math behind calculating fair scoring needs to be revised.
Can you explain the key points of what makes this bad? I provided 2 versions of the formula in the earlier posts and OP decided to mend them together. What kind of behavior are you expecting due to the results of a team match?

The reasoning behind the formulas can be found in the earlier post: p/4921364


First of all, how is ratio ever going to be > 1 if the ratio is calculated by achieved score/max possible score?

2. I can break this formula so easily. I just put on nofail and get a score of 0. What happens to your formula then?

ln(-.01)^-1 = a complex number. How are you going to calculate imaginary numbers? Is the score going to have an imaginary component and a real component?

Let's take ln(-.01)^-1 for example. Wolfram is telling me the answer is -1.48185 - 1.0109i.

This implies we need to calculate 1.794e^(i*34.3degrees + pi/2)

Have fun with that.

Your formula needs work.
abraker

Minsoo wrote:

First of all, how is ratio ever going to be > 1 if the ratio is calculated by achieved score/max possible score?
I should have put subscripts to those ratios. The ratio in the formula for s and the formula for Play Level are different. I simplified the formula for s which is actually based on these. Sorry for the confusion. I'll try to get the OP to update it with the actual variables replacing "ratio" when I get the chance.




Minsoo wrote:

2. I can break this formula so easily. I just put on nofail and get a score of 0. What happens to your formula then?
ln(-.01)^-1 = a complex number. How are you going to calculate imaginary numbers? Is the score going to have an imaginary component and a real component?
Let's take ln(-.01)^-1 for example. Wolfram is telling me the answer is -1.48185 - 1.0109i.
This implies we need to calculate 1.794e^(i*34.3degrees + pi/2)
Have fun with that.
A quick fix would be to clip it to the -258.2 points minimum. I would only hope that wouldn't cause too much inflation within the sum of all Team Points (I'm basing that typical >1 team score ratios would be generally be no more than 100:1, raising the sum all Team Points by about 1 point for such plays). In an event of 1000:1 ratio plays, I expect the total sum to rise by about a bit less than 200, which might be a good or bad thing depending on the rate at which player accounts having greater than 200 team points become inactive.

Ideally, I want to balance out the <1.0 and >1.0 score ratio results though so that there is no inflation and no deflation within the system.
Minsoo
The problem with "clipping" things -- or in general making it so that if the formula hits a specific barrier it does something instead... is that then the formula doesn't work for all situations. Also again, if a team decides to all put on no fail and get 0s, ln(0) = -infinity



You can't have a "if there is a big gap, use this formula, else use this formula".

This puts processing power to waste and also just mathematically does not look good.

You're also scaling it a bit weirdly, instead why not scale things with accuracy as well? I'm not entirely sure what you mean by "diff" (difficulty? -> because difficulty * accuracy doesn't make much sense...)

Something that would make a bit more sense with your formula would be

teamA = starDifficulty * Math.log1p(Math.e + 100*acc) * Math.log1p(Math.e + avgA)

and then you scale their points based off of teamA / total, and whatnot.
abraker

Minsoo wrote:

The problem with "clipping" things -- or in general making it so that if the formula hits a specific barrier it does something instead... is that then the formula doesn't work for all situations. Also again, if a team decides to all put on no fail and get 0s, ln(0) = -infinity
The clipping is supposed to fix the -infinity by not allowing the ratio to be less than or equal to 0.01. I am not sure what you mean when yousay it doesn't work for all situations.

Minsoo wrote:

You can't have a "if there is a big gap, use this formula, else use this formula".
This puts processing power to waste and also just mathematically does not look good.
That was my solution to what the OP wanted. OP wanted both, so I made it to have both. While I like the big gap formula better, I think the unpredictability factor in the small gap formula is not bad either.

Minsoo wrote:

You're also scaling it a bit weirdly, instead why not scale things with accuracy as well? I'm not entirely sure what you mean by "diff" (difficulty? -> because difficulty * accuracy doesn't make much sense...) and then you scale their points based off of teamA / total, and whatnot.
I don't use accuracy because it is the score part that is supposed to be equivalent to the measurement of the cumulative performance up to a point in time on a map. Also score is already influenced by accuracy, albeit much less than combo.
Minsoo
Either way, log(ratio) of whatever wouldn't be very good to use as a scaling factor because it varies easily and dramatically. Especially when it behaves near log(1).
abraker

Minsoo wrote:

Either way, log(ratio) of whatever wouldn't be very good to use as a scaling factor because it varies easily and dramatically. Especially when it behaves near log(1).
It is because of how it behaves I am using it. The reciprocal of a value (the ratio) would result in the negative equivalent result. Using this, it is possible to reward the winning team N number of points and the loosing team -N number of points logarithmically.
Topic Starter
Arrcival

abraker wrote:

That was my solution to what the OP wanted. OP wanted both, so I made it to have both. While I like the big gap formula better, I think the unpredictability factor in the small gap formula is not bad either.
Yeah, both isn't a good idea?

abraker wrote:

Minsoo wrote:

You're also scaling it a bit weirdly, instead why not scale things with accuracy as well? I'm not entirely sure what you mean by "diff" (difficulty? -> because difficulty * accuracy doesn't make much sense...) and then you scale their points based off of teamA / total, and whatnot.
I don't use accuracy because it is the score part that is supposed to be equivalent to the measurement of the cumulative performance up to a point in time on a map. Also score is already influenced by accuracy, albeit much less than combo.
Score v2 can be an idea, but idk if it's good for the formulas ^^'
Illya X Ram Li
I'm agree :d
Topic Starter
Arrcival
Thanks you...
But if u guys have any idea or correction please say! :p
EnDy_S
Still waiting for this one :( Any news?
Frozen_Rhythm
Obviously everyone loves it.
This feature needs a lot of work.
We will wait for your kind response peppy!
Topic Starter
Arrcival

EnDy_S wrote:

Still waiting for this one :( Any news?
I'm waiting too :(
No news for the moment...

Frozen_Rhythm wrote:

Obviously everyone loves it.
This feature needs a lot of work.
We will wait for your kind response peppy!
Thanks :p
I wait for peppy too, why not contacting me? x')
show more
Please sign in to reply.

New reply