forum

Store Best Performance apart from Highest Scores in Map Rank

posted
Total Posts
25
This is a feature request. Feature requests can be voted up by supporters.
Current Priority: +7
Topic Starter
Full Tablet
Currently, only the play that has the highest score is chosen for pp calculations. When choosing what play is considered for pp calculations, the best play isn't always chosen because of this.

Examples of when this can happen:

1) Player A gets 6 million score in a map with 95% accuracy, then the same player plays the map again and gets 5.6 million score with only 90% accuracy (resulting in less pp because of less accuracy). So in the end, the second play results in a penalty in pp score. This goes against the statement included in the wiki that there is no penalty for bad scores. Another similar case is getting a high-score in a map with bad performance, that makes other better plays with less score don't count.

2) Player B gets a rank 1 in a map with a 4-mod SSH, getting a good amount of pp. Then, let's say 2 years later, the pp gained in that map practically banishes. Now player B won't be able to get pp from that map again because he can't get a higher score in that map, even though he can achieve the same or slightly less score (that deserves pp based on the performance), another player then would get more (unweighted) pp than the best player with a worse performance even if they played the map the same day. No longer valid since there is no performance decay in ppv2.

One possible solution to this issue is adding new entries in the server database that store the plays with best performance in each song (each entry has the same information as the score entries, such as max combo, judgments, etc...). This could add some server-load, but I believe it is a step necessary to fully implement the pp system to the game.

Also, the new entries could be used to generate 'Performance Rankings' for each map, so we can know more easily which player had the best calculated performance in the maps (and not only who scored more).

As for changes in the pp calculation algorithm that can make the performance values change (making the best performance rankings for maps change, and possibly making the current best performance for each player stored not be the best performance ever achieved by the player anymore): changes in the algorithm usually don't change the values so much (usually, after an algorithm change, a DT FC score will remain being considered a better performance than a HR+HD play the same player did, if the DT FC was considered better before the algorithm change), keeping score as the defining factor for what play is kept for performance calculation causes more incongruencies than if they kept best performances and then their values change.
Edit: made it more relevant to ppv2 instead of ppv1.
chaotic_iak
Bumping this thread, because I'm thinking of the same thing too (tl;dr: I support). All my points have been addressed above.

Although here's some point so this post is not an empty "support" post:

Full Tablet wrote:

Player B gets a rank 1 in a map with a 4-mod SSH, getting a good amount of pp. Then, let's say 2 years later, the pp gained in that map practically banishes.
If the "pp decay" is exponential instead of linear (all decays, in-game or otherwise, should be exponential anyway), at 2 years the pp gain becomes 1/4 of the original score. I thought about this instead:
Suppose you get 200 pp from a 4-mod SS. After a year, it becomes 100 pp, and you decide to play the map again. As long as you get the same score--heck, I think even a less score as long as it gives more than 100 pp normally is sufficient--you should have still proven that your performance is still "great" and get back the lost pp.
ryza
Doesn't seem particularly relevant to me
If you got good acc once, you can do it again
chaotic_iak
The problem is that by simply matching the score you cannot regain your pp since the "best" score is the earlier one. Suppose in a no-spinner map you get total perfection (SS with 4 mods). You will not be able to replenish your pp from that song any more.
ryza
Then really, play other maps

it in no way matters after a year (or more)

Your performance on the map is denoted by the best score kept in osu's database, if you have worse acc while playing and are really so concerned about losing pp from a higher score, then quit out before you finish the song.

The difference in pp is mostly negligible anyways

also I don't really know how peppy's calculations work, but if you have a 4mod SS, from my observations it doesn't seem like that loses very much value over time (or is very likely not giving you much in the first place)
Topic Starter
Full Tablet

chaotic_iak wrote:

Bumping this thread, because I'm thinking of the same thing too (tl;dr: I support). All my points have been addressed above.

Although here's some point so this post is not an empty "support" post:

Full Tablet wrote:

Player B gets a rank 1 in a map with a 4-mod SSH, getting a good amount of pp. Then, let's say 2 years later, the pp gained in that map practically banishes.
If the "pp decay" is exponential instead of linear (all decays, in-game or otherwise, should be exponential anyway), at 2 years the pp gain becomes 1/4 of the original score. I thought about this instead:
Suppose you get 200 pp from a 4-mod SS. After a year, it becomes 100 pp, and you decide to play the map again. As long as you get the same score--heck, I think even a less score as long as it gives more than 100 pp normally is sufficient--you should have still proven that your performance is still "great" and get back the lost pp.
The pp practically banishes since the pp each map gives is weighted based on the position inside your best performances list.
For example: You get 1200 raw pp for your best performance (and it gives you 1200 pp, since it is #1 in your performance list), then 2 years later it gets down to 300 raw pp. The 300 raw pp will likely become 0 pp for you since you will have several scores with around 1000 raw pp, so the score with 300 pp before weighting will be, for example, place #4000 in the list.

Noticing how fast the decay is would be hard for a player that has many plays, since the position of the beatmaps in the top performance list will constantly change in a way the pp lost in the decay is dampened by the presence of more recently played maps. (Notice this might happen even if the player suddenly stops playing completely).

(The best way to measure how fast the decay is observing the pp of a player that has exactly 10 beatmaps played with good pp, all of them in the same day).

Silynn wrote:

Then really, play other maps

it in no way matters after a year (or more)

Your performance on the map is denoted by the best score kept in osu's database, if you have worse acc while playing and are really so concerned about losing pp from a higher score, then quit out before you finish the song.

The difference in pp is mostly negligible anyways

also I don't really know how peppy's calculations work, but if you have a 4mod SS, from my observations it doesn't seem like that loses very much value over time (or is very likely not giving you much in the first place)
A 4mod SS in a rewarding map (for example a very popular and contested easy 5 star song) would give a good amount of pp. About how fast these records decay over time, we would need to measure it (that might take some time, and the most accurate and fastest way would require multi-accounting which is illegal), or wait until peppy gives us the exact formulas.

About quitting before finishing the map, there are 2 issues with that:

- We don't always really know if a current play is better than the recorded one. For example: One has a SS on certain difficult map, then, when playing with DT, with 92% acc and FC (and only a few circles before finishing the song), you can't be sure if it is convenient to record the current score or not. You can't really know if the better score rank bonus will overcome the accuracy penalty in that case (specially if the map is really hard to play with mods in the first place, since there won't be many people that play with them in the score ranking, beating SS scores).

- (This may not apply to all players) Having to worry about if your current play is better than the one you recorded is too distracting. Concentrating in the beatmap is be hard if you have to evaluate how well you are doing, and there is and added nervousness because there is a possibility you screw up one of your best performances.
TheVileOne
Shouldn't pp be based on highest combo given as well as accuracy attempted?

All of this could go into an exponential equation.

(Accuracy points+ rank points + mod points) to the power of 1.xxxx where xxxx = max combo achieved. 1000 combos = 1.1, 585 = 1.0585, 32 = 1.0032 etc

Lets say adding a mod will give you 3 pp per mod. Lets ignore variances for the sake of simplicity.

Accuracy could give 10 to the power of x. x = accuracy 100 = 1.0, 99 = 0.99

Lets say rank points are fixed at 5, less than accuracy but more than mod points.

Lets say a map has 600 combos, and there is a person named Bob FCs and gets a middle ranged rank of 200 with no mods. His accuracy is 88%. By my formula, this would equate to.

(0+5+10 x (10x0.88) )^1.06 = 14.65118623 pp

Lets add a mod to it, and Bob improves his score to rank 167, but his accuracy drops to 84%. Lets say he got 592 combos.

(3+5+10 x (10x0.84))^1.0592 = 17.50668109 pp +3 pp even with less accuracy and less combos. Lets say the mod powerup didn't give more pp. Answer = 13.80150807 -1 pp.

What peppy needs to do is to find a balance between these values. When you add a mod, the lowest combo/accuracy needed to beat a score without that mod needs to be considered as a technical higher performance. Basically the numbers need to indicate this:

No mod x combo with y accuracy = z score (Although probably would be accurate enough in most cases, higher combo = higher score, the alternative would be dynamic point calculating. Basically takes the info from each point in the song and calculate score in between misses, and then compare that to a mod. This data could easily become very complex and hard to track. Mod, accuracy, and max combo should be sufficient in most cases. (As well as rank)).

What combo with what accuracy is needed to obtained the same score with 1 mod, with 2,3,4 etc

Back to my equation. What values could mod be to make the output pp = 14 or more?

Mod = 1 pp pp = 15.03104134

Equation P(x) = 14.65118623
(x+5+10(10x0.84))^1.0592
(x+5+6.918309709)^1.0592
(x+11.918309709)^1.0592 = 14.65118623

P(0.8) = 14.78466937
P(0.7) = 14.66156914
P(0.6955) = 14.65603098
P(0.693) = 14.65295428
P(0.6915) = 14.65110828
P(0.69155) = 14.65116981
P(0.691555) = 14.65117596
P(0.69156) = 14.6511.8212
P(0.6915625) = 14.65118519
P(0.6915634) = 14.6511863
P(0.69156334) = 14.65118623 approx

If the mod was worth this, with that accuracy, it would be worth approximately the same pp. If this is worth the same pp, then if the player had gotten a FC it would give 14.68092288 pp. If it was a SS then it would be equal 18.50999013 with a mod, and 17.646425 without a mod. Not that much, but the higher the value given to the mod part, the greater the difference in performance will be.

From what I hear the SS bonus imbalances pp gain, because it is worth more than the pp given by adding a mod or going up a rank. This should be balanced in the algorithm. Under no circumstance should playing with a harder mod give you less pp, because the point scales rise exponentially or should rise exponentially) so that the SS of mod count-1 will always be less than or equal to the equivalent score with mods applied.

Edit: I forgot to mention that ensuring exponential scaling is done by adding a variable to the value of an exponent

10 to the 0.58+0.05 0.05 = mod pp increase factor
Topic Starter
Full Tablet

TheVileOne wrote:

Shouldn't pp be based on highest combo given as well as accuracy attempted?

All of this could go into an exponential equation.

(Accuracy points+ rank points + mod points) to the power of 1.xxxx where xxxx = max combo achieved. 1000 combos = 1.1, 585 = 1.0585, 32 = 1.0032 etc

Lets say adding a mod will give you 3 pp per mod. Lets ignore variances for the sake of simplicity.

Accuracy could give 10 to the power of x. x = accuracy 100 = 1.0, 99 = 0.99

Lets say rank points are fixed at 5, less than accuracy but more than mod points.

Lets say a map has 600 combos, and there is a person named Bob FCs and gets a middle ranged rank of 200 with no mods. His accuracy is 88%. By my formula, this would equate to.

(0+5+10 x (10x0.88) ) to the 1.06 power = 14.65118623 pp

Lets add a mod to it, and Bob improves his score to rank 167, but his accuracy drops to 84%. Lets say he got 592 combos.

(3+5+10 x (10x0.84)) to the 1.0592 power = 17.50668109 pp +3 pp even with less accuracy and less combos. Lets say the mod powerup didn't give more pp. Answer = 13.80150807 -1 pp.
Increasing the exponent linearly with combo isn't a very numerically stable way to calculate score.
Example: http://osu.ppy.sh/b/156352&m=0 Uan has 7261 Combo, 99.14% acc with no mods.
(0+5+10*10*0.9914)^1.7261 = 3038,15
Doesn't it seem to be too much?
Edit: the notation was confusing.
(0+5+10^0.9914)^1.7261=209,516 104,758 (It seems I made a typo in the calculator)
TheVileOne
Ummm

0+5+10^0.9914 = 14.80392545

14.80392545^1.7261 = 104.7580845 pp

I don't know where you're getting these high numbers. (Exponents come before addition). If the player got 10,000 combo, that would be 15^2 = 225 pp without mods SS run.

Edit: Updated with proper notation.
Edit: More info

If you took that raw pp and then multiplied it exponentially by 1-c-a or maybe just multiply it normally by a really low number based on 1-c-a. Both would work

c= contest status and a = average performance factors. Both will be less than 1, because they will reduce how much pp overall is awarded by you. The more highly contested a map is, the lower the value of c and the value of a will be values compared to the average performance. Higher than average = lower value of a and lower means a higher value of a.

Example c = 0.3 and a = 0.6 1-0.3-0.6 = 0.1 Final pp given to a player under these stats will be 10% of raw pp given by algorithm
Topic Starter
Full Tablet

TheVileOne wrote:

Ummm

0+5+10^0.9914 = 14.80392545

14.80392545^1.7261 = 104.7580845 pp

I don't know where you're getting these high numbers. (Exponents come before addition). If the player got 10,000 combo, that would be 15^2 = 225 pp without mods SS run.
That numbers seems to be too high if you included it to the current weighting system (that weights pp based on the position on your top performance list) (and if you say we should ignore it, that would make pp only another score ranking, absolutely farmable). Currently, raw pp scores (before weighting) are in the range of several hundreds (I estimate that 300 is a newbie top score, and 1500 is something a pro player would get in a song, but maybe I am very far from the real numbers).
With your formula:
http://osu.ppy.sh/b/95875?m=0 White Wolf. 100%Acc, 922 Combo, 2 mods (probably the most pp rewarding play currently).
(6+5+10)^1.0922= 27.805. Let's scale it to 1600.
Then Uan would have 6028.15 pp from his Unforgiving play only.
Edit: Edited all decimals numbers to use the USA standard for points and commas.
TheVileOne
I left out weights from this equation to simplify it. After you get raw pp, apply weight to it, thus reducing score. Weight = x% of pp. Also I'm not interested in the current values of pp. Under my equation, no 1 song without mods would give more than 225 pp. This total amount can be chipped away by contested, and average performance and whatever else. The numbers weren't even that important, as this algorithm I made is to stand as a model of the pp system. It needs to accurately determine performance, and high combo = harder difficulty usually/ marathon which is also hard to FC. <Also the circle to slider ratio is an indicator of difficulty. I didn't weigh in variable pp by rank, which would obviously change the value of Rank variable (the 5) to other values.

This is not intended to be a replacement for the pp system as it sits, but more as a template for how the pp system probably works. It's obvious to me that there are serious issues with how the current algorithm determines performance and my 80% complete algorithm is already more accurate at determining pp. I just left out the factors that I do not have a clear idea on how they are calculated. These can be added to this template at any time.

Whatever you're doing to calculate exponents. you're doing it incorrectly. None of my exponents are greater than 2. And the maximum input valus maxes out at 15 without mods, and without variable rank.

Max pp attainable with all 4 mods if we say each mod is worth 3.

(12+5+10^1)^1.0920

(27)^1.0920 = 36.56362368 pp

The square root of 27,805 is 166.7483133 (A max root of 27 will not get this value)
Topic Starter
Full Tablet

TheVileOne wrote:

I left out weights from this equation to simplify it. After you get raw pp, apply weight to it, thus reducing score. Weight = x% of pp. Also I'm not interested in the current values of pp. Under my equation, no 1 song without mods would give more than 225 pp. This total amount can be chipped away by contested, and average performance and whatever else. The numbers weren't even that important, as this algorithm I made is to stand as a model of the pp system. It needs to accurately determine performance, and high combo = harder difficulty usually/ marathon which is also hard to FC. <Also the circle to slider ratio is an indicator of difficulty. I didn't weigh in variable pp by rank, which would obviously change the value of Rank variable (the 5) to other values.

This is not intended to be a replacement for the pp system as it sits, but more as a template for how the pp system probably works. It's obvious to me that there are serious issues with how the current algorithm determines performance and my 80% complete algorithm is already more accurate at determining pp. I just left out the factors that I do not have a clear idea on how they are calculated. These can be added to this template at any time.

Whatever you're doing to calculate exponents. you're doing it incorrectly. None of my exponents are greater than 2. And the maximum input valus maxes out at 15 without mods, and without variable rank.

Max pp attainable with all 4 mods if we say each mod is worth 3.

(12+5+10^1)^1.0920

(27)^1.0920 = 36.56362368 pp

The square root of 27,805 is 166.7483133 (A max root of 27 will not get this value)
Sorry, I mixed up dot and commas somewhere there.

What is arguing here is that your formula is blind when it comes to calculate combo bonus. Currently as it is, most songs (that have a not so incredibly high max combo) have a combo bonus that is roughly linear (roughly combo*basescore*Ln(basescore)/10000 bonus), and when it comes to incredibly long songs, it becomes roughly quadratic (combo*basescore^2*Ln(basescore)/10000) (and in hypothetic long songs, it can surpass quadratic bonus). (the rough formulas I gave are the score given per each circle when doing the combo, basescore is Rank+Mod+Acc).

I think most players would have a much harder time FCing a 50 circle beatmap that is like a DT Freedom Dive [4D], than getting a high combo in The Unforgiving (the main difficulty in that song is dealing with boredom, about 10 seconds in that map are at least remotely difficult for a decent player, you can easily get 2000 combo or more if you don't asleep).

Also, what's your opinion about the feature I proposed in this thread? (Storing top pp scores in another database to be able to determine best performance based on it instead of just using the top score to calculate pp in each map).
TheVileOne
I think the pp system should and can be fixed, by weighting max combo and mods used over accuracy. pp is related to rank and accuracy. Rank needs to outweigh accuracy, because a player's score is compared to other people, not accuracy. What determines score achieved? Mods, used, and max combos attained. I would rather they fix the pp system rather than provide a way to work around the fixable flaws in the system.
[Dellirium]
I want to bump this topic because mine was duplicated.
[Dellirium]
Oh come on.
Topic Starter
Full Tablet
Has been this considered for implementation?

Is there a reason this haven't been implemented? (For example: not enough resources, this is incompatible with the current pp system design, this can be a bad idea somehow, it is just low priority,etc...)
ryza
It's a bad idea
Topic Starter
Full Tablet

Silynn wrote:

It's a bad idea
Can you explain, please?
ryza
For one, stop bitching about useless stuff

and second, get better at osu

Seriously, if you are at a point where you are actually getting pp for a rank 300, one, you shouldn't even care about your pp amount, and two, you also probably don't know what you are talking about

I spend a very large amount of time taking advantage of this system and I can tell you that if you are losing points for increasing your /rank/ on a song (which is what increases your PP multiplier the most - rank 1 is sometimes double the points of rank 2 and so on) then you likely don't have enough performance points for any changes to the system to be relevant or meaningful in any way

and 4 mod SS? That must be a normal or an easy (in some rare cases a hard), and it very likely isn't giving you performance points anyways. And if you play a decent bit at all in the VERY long time it takes for those points to decay (I don't even know how decay works - I've seen very old scores still giving people lots of pp) - you will easily gain the points back over the time it takes for the decay to happen (I don't see the point in bitching about the decay when you can't even beat the score from a year ago and it's completely intended to happen)

There's a lot more things I can say, but it all comes down to this:

what you are asking for is something horribly irrelevant and not meaningful at all and you should just get better at the game before even deciding to care about performance points in general
Jenny
^

the day you gain pp is the day you don't give craps about them anymore but try to get good/perfect scores on the maps you enjoy (or rather, that is how it should be - just play what you want and like you want; if you gain pp, okay, if not, okay - why care about that number?)

  • omg but graevd maps a de besd - mod and spread them around then, because that is how shit (speaking of maps in general here) gets ranked, by people's attention.
Topic Starter
Full Tablet
Well, so the pp system should only focus in ranking people that are really good at the game?
I think the aim of the pp system is to create an accurate ranking system that measures as much players as possible. The fact that the system is perceived as inaccurate when rankings players (specially players with low pp ranking) is something that should be corrected (and things seem to be moving that way; in the last weeks, I haven't seen many people that have a ranking near mine that have a substantial difference in skill, for example).

With this request, some of the issues with the pp system can be solved.

The pp a score gives is determined by various factors, one of them is score rank. At your current pp rank, score rank is probably perceived as the most important factor since each pp-giving score already has a high value in all the factors, and the score rank is the most variable factor of them (for example, 99.87% acc vs 99.44% acc can possibly mark a difference between #5 and #12 in some beatmaps, comparatively there is a bigger difference in score ranking than in acc). The problem mentioned in this thread doesn't happen when score rank in a map is the only thing that makes a difference in the amount of pp the map gives, but for most players score rank is not the only factor that seems to make a difference.

Silynn wrote:

[...] (I don't see the point in bitching about the decay when you can't even beat the score from a year ago and it's completely intended to happen)[...]
Why is it intended to happen?
If player A did a score with a performance of 100 "points", then, after the score decays to 90 "points", a performance of 95 "points" should be ignored? And another player B that achieves 95 "points" should get more points than player A for the same map, even if both achieved 95 "points" the same day?
ryza
If it wasn't intended to happen, then decay wouldn't exist in the first place.

If you get a score and can't improve it, then you don't deserve any more PP than what it is giving you (even if you lost PP for lower accuracy - which is likely to almost never happen)

If you really want more PP from it, then by all means go ahead and improve it. Simple as that.
Jenny

Full Tablet wrote:

At your current pp rank, score rank is probably perceived as the most important factor since each pp-giving score already has a high value in all the factors, and the score rank is the most variable factor of them (for example, 99.87% acc vs 99.44% acc can possibly mark a difference between #5 and #12 in some beatmaps, comparatively there is a bigger difference in score ranking than in acc).
Rather than seing it as 99.87% and 99.44% accuracy, you should perceive it as the 99.44% guy having screwed up three times as much as the 99.87% one - the 99.87% score got 1 100, the 99.44% probably got 3, so there you go, in terms of 100s (-> messed up hits), he is thrice as bad, thus, the difference.

Viewing and perceiving it like this, the accuracy value in the pp system makes more sense (imo).
Topic Starter
Full Tablet

Silynn wrote:

If it wasn't intended to happen, then decay wouldn't exist in the first place.

If you get a score and can't improve it, then you don't deserve any more PP than what it is giving you (even if you lost PP for lower accuracy - which is likely to almost never happen)

If you really want more PP from it, then by all means go ahead and improve it. Simple as that.
I think the decay is intended to reduce pp for people that become inactive or for any reason become worse at the game, so the pp rankings reflect better the current situation of the players. For that reason, the decay shouldn't punish to a point where it undervalues the current skill of the player.

Here is an analogy: There is a basketball player that is well known to have near 100% accuracy rate when throwing the ball for 3 points. Years later, the player starts getting older and thus his accuracy rate is near 90% with the same kind of throws. Now, the fact that the player has become worse than before doesn't mean he is currently worse than any other player that also has 90% accuracy with the throw.

Jenny wrote:

Rather than seing it as 99.87% and 99.44% accuracy, you should perceive it as the 99.44% guy having screwed up three times as much as the 99.87% one - the 99.87% score got 1 100, the 99.44% probably got 3, so there you go, in terms of 100s (-> messed up hits), he is thrice as bad, thus, the difference.

Viewing and perceiving it like this, the accuracy value in the pp system makes more sense (imo).
Well, that makes sense. But that means that a SS (or SSH) is infinite times as good as any other score?

Knowing the exact formulas for pp calculation would make this discussion much easier (and would give us more chances to discuss about the core of the pp system). With the formulas we could know what acc is "x" times better than other acc (according to pp calculations).
Jenny
Well, as peppy stated some time ago, the closer you get to an SS(H), the pp you gain gets exponentially better - I remember lewa getting more pp of a rank 2 HRHD SS than from getting rank 1 on the same map on DTHD with a bit less accuracy later on, as an example; a flawless play will grant you a big advantage as it shows you're doing perfectly at what you are doing, rather than "just good" at something else (-> HRHD SS instead of DTHD S)
Please sign in to reply.

New reply