h1or4dme8/osu-lazer
1
0
Fork 0
mirror of https://github.com/ppy/osu.git synced 2025-03-13 12:07:19 +08:00
Code Issues Releases Wiki Activity

Massively improve estimation accuracy using a folded distribution

Browse Source
This commit is contained in:
Natelytle 2023-02-17 00:28:09 -05:00
parent db0fda0638
commit 2d7c99b6f3
1 changed files with 17 additions and 14 deletions

31
osu.Game.Rulesets.Mania/Difficulty/ManiaPerformanceCalculator.cs

@ -7,6 +7,7 @@ using System;
using System.Collections.Generic;
using System.Linq;
using MathNet.Numerics;
using MathNet.Numerics.Distributions;
using osu.Framework.Audio.Track;
using osu.Framework.Extensions.IEnumerableExtensions;
using osu.Game.Rulesets.Difficulty;
@ -109,27 +110,24 @@ namespace osu.Game.Rulesets.Mania.Difficulty
double p50Note = hitProb(h50, d) - hitProb(h100, d);
double p0Note = 1 - hitProb(h50, d);
// Effective hit window for LN tails. Should be a value between 1 and 2. This is because the hit window for LN tails in stable
// arent static, and depend on how far from 0ms offset the hit on the head was. A lower value results in a lower estimated deviation.
const double tail_multiplier = 1.5;
// Since long notes only give a specific judgement if both both hits end up within a certain hit window,
// multiply the probability of hitting in the head hit window by the probability of hitting in the tail hit window.
double pMaxLn = hitProb(hMax * 1.2, d) * hitProb(hMax * 1.2 * tail_multiplier, d);
// Since legacy LN tails take the absolute error of both hit judgements on an LN, we need to use a folded normal distribution to calculate it.
double pMaxLn = hitProb(hMax * 1.2, d) * hitProbLn(hMax * 2.4, d);
double p300Ln = hitProb(h300 * 1.1, d) * hitProb(h300 * 1.1 * tail_multiplier, d)
- hitProb(hMax * 1.2, d) * hitProb(hMax * 1.2 * tail_multiplier, d);
double p300Ln = hitProb(h300 * 1.1, d) * hitProbLn(h300 * 2.2, d)
- hitProb(hMax * 1.2, d) * hitProbLn(hMax * 2.4, d);
double p200Ln = hitProb(h200, d) * hitProb(h200 * tail_multiplier, d)
- hitProb(h300 * 1.1, d) * hitProb(h300 * 1.1 * tail_multiplier, d);
double p200Ln = hitProb(h200, d) * hitProbLn(h200 * 2, d)
- hitProb(h300 * 1.1, d) * hitProbLn(h300 * 2.2, d);
double p100Ln = hitProb(h100, d) * hitProb(h100 * tail_multiplier, d)
- hitProb(h200, d) * hitProb(h200 * tail_multiplier, d);
double p100Ln = hitProb(h100, d) * hitProbLn(h100 * 2, d)
- hitProb(h200, d) * hitProbLn(h200 * 2, d);
double p50Ln = hitProb(h50, d) * hitProb(h50 * tail_multiplier, d)
- hitProb(h100, d) * hitProb(h100 * tail_multiplier, d);
double p50Ln = hitProb(h50, d) * hitProbLn(h50 * 2, d)
- hitProb(h100, d) * hitProbLn(h100 * 2, d);
double p0Ln = 1 - hitProb(h50, d) * hitProb(h50 * tail_multiplier, d);
double p0Ln = 1 - hitProb(h50, d) * hitProbLn(h50 * 2, d);
double pMax = ((pMaxNote * attributes.NoteCount) + (pMaxLn * attributes.HoldNoteCount)) / totalHits;
double p300 = ((p300Note * attributes.NoteCount) + (p300Ln * attributes.HoldNoteCount)) / totalHits;
@ -246,5 +244,10 @@ namespace osu.Game.Rulesets.Mania.Difficulty
{
return SpecialFunctions.Erf(x / (deviation * Math.Sqrt(2)));
}
private double hitProbLn(double x, double deviation)
{
return Math.Pow(2 * Normal.CDF(0, deviation * Math.Sqrt(2), x) - 1, 2);
}
}
}