1
0
mirror of https://github.com/ppy/osu.git synced 2024-11-15 17:57:29 +08:00
osu-lazer/osu.Game/Database/StandardisedScoreMigrationTools.cs
Bartłomiej Dach 45fcbea182
Compute total score without mods during standardised score conversion
This is going to be used by server-side flows. Note that the server-side
overload of `UpdateFromLegacy()` was not calling
`LegacyScoreDecoder.PopulateTotalScoreWithoutMods()`.

Computing the score without mods inline reduces reflection overheads
from constructing mod instances, which feels pretty important for
server-side flows.

There is one weird kink in the treatment of stable scores with score V2
active - they get the *legacy* multipliers unapplied for them because
that made the most sense. For all intents and purposes this matters
mostly for client-side replays with score V2. I'm not sure whether
scores with SV2 ever make it to submission in stable.

There may be minute differences in converted score due to rounding
shenanigans but I don't think it's worth doing a reverify for this.
2024-05-21 13:11:08 +02:00

700 lines
37 KiB
C#

// Copyright (c) ppy Pty Ltd <contact@ppy.sh>. Licensed under the MIT Licence.
// See the LICENCE file in the repository root for full licence text.
using System;
using System.Collections.Generic;
using System.Diagnostics;
using System.Linq;
using osu.Framework.Logging;
using osu.Game.Beatmaps;
using osu.Game.Extensions;
using osu.Game.IO.Legacy;
using osu.Game.Rulesets;
using osu.Game.Rulesets.Judgements;
using osu.Game.Rulesets.Mods;
using osu.Game.Rulesets.Objects;
using osu.Game.Rulesets.Scoring;
using osu.Game.Rulesets.Scoring.Legacy;
using osu.Game.Scoring;
namespace osu.Game.Database
{
public static class StandardisedScoreMigrationTools
{
public static bool ShouldMigrateToNewStandardised(ScoreInfo score)
{
if (score.IsLegacyScore)
return false;
if (score.TotalScoreVersion > 30000002)
return false;
// Recalculate the old-style standardised score to see if this was an old lazer score.
bool oldScoreMatchesExpectations = GetOldStandardised(score) == score.TotalScore;
// Some older scores don't have correct statistics populated, so let's give them benefit of doubt.
bool scoreIsVeryOld = score.Date < new DateTime(2023, 1, 1, 0, 0, 0);
return oldScoreMatchesExpectations || scoreIsVeryOld;
}
public static long GetNewStandardised(ScoreInfo score)
{
int maxJudgementIndex = 0;
// Avoid retrieving from realm inside loops.
int maxCombo = score.MaxCombo;
var ruleset = score.Ruleset.CreateInstance();
var processor = ruleset.CreateScoreProcessor();
processor.TrackHitEvents = false;
var beatmap = new Beatmap();
HitResult maxRulesetJudgement = ruleset.GetHitResults().First().result;
// This is a list of all results, ordered from best to worst.
// We are constructing a "best possible" score from the statistics provided because it's the best we can do.
List<HitResult> sortedHits = score.Statistics
.Where(kvp => kvp.Key.AffectsCombo())
.OrderByDescending(kvp => processor.GetBaseScoreForResult(kvp.Key))
.SelectMany(kvp => Enumerable.Repeat(kvp.Key, kvp.Value))
.ToList();
// Attempt to use maximum statistics from the database.
var maximumJudgements = score.MaximumStatistics
.Where(kvp => kvp.Key.AffectsCombo())
.OrderByDescending(kvp => processor.GetBaseScoreForResult(kvp.Key))
.SelectMany(kvp => Enumerable.Repeat(new FakeJudgement(kvp.Key), kvp.Value))
.ToList();
// Some older scores may not have maximum statistics populated correctly.
// In this case we need to fill them with best-known-defaults.
if (maximumJudgements.Count != sortedHits.Count)
{
maximumJudgements = sortedHits
.Select(r => new FakeJudgement(getMaxJudgementFor(r, maxRulesetJudgement)))
.ToList();
}
// This is required to get the correct maximum combo portion.
foreach (var judgement in maximumJudgements)
beatmap.HitObjects.Add(new FakeHit(judgement));
processor.ApplyBeatmap(beatmap);
processor.Mods.Value = score.Mods;
// Insert all misses into a queue.
// These will be nibbled at whenever we need to reset the combo.
Queue<HitResult> misses = new Queue<HitResult>(score.Statistics
.Where(kvp => kvp.Key == HitResult.Miss || kvp.Key == HitResult.LargeTickMiss)
.SelectMany(kvp => Enumerable.Repeat(kvp.Key, kvp.Value)));
foreach (var result in sortedHits)
{
// For the main part of this loop, ignore all misses, as they will be inserted from the queue.
if (result == HitResult.Miss || result == HitResult.LargeTickMiss)
continue;
// Reset combo if required.
if (processor.Combo.Value == maxCombo)
insertMiss();
processor.ApplyResult(new JudgementResult(null!, maximumJudgements[maxJudgementIndex++])
{
Type = result
});
}
// Ensure we haven't forgotten any misses.
while (misses.Count > 0)
insertMiss();
var bonusHits = score.Statistics
.Where(kvp => kvp.Key.IsBonus())
.SelectMany(kvp => Enumerable.Repeat(kvp.Key, kvp.Value));
foreach (var result in bonusHits)
processor.ApplyResult(new JudgementResult(null!, new FakeJudgement(result)) { Type = result });
// Not true for all scores for whatever reason. Oh well.
// Debug.Assert(processor.HighestCombo.Value == score.MaxCombo);
return processor.TotalScore.Value;
void insertMiss()
{
if (misses.Count > 0)
{
processor.ApplyResult(new JudgementResult(null!, maximumJudgements[maxJudgementIndex++])
{
Type = misses.Dequeue(),
});
}
else
{
// We ran out of misses. But we can't let max combo increase beyond the known value,
// so let's forge a miss.
processor.ApplyResult(new JudgementResult(null!, new FakeJudgement(getMaxJudgementFor(HitResult.Miss, maxRulesetJudgement)))
{
Type = HitResult.Miss,
});
}
}
}
private static HitResult getMaxJudgementFor(HitResult hitResult, HitResult max)
{
switch (hitResult)
{
case HitResult.Miss:
case HitResult.Meh:
case HitResult.Ok:
case HitResult.Good:
case HitResult.Great:
case HitResult.Perfect:
return max;
case HitResult.SmallTickMiss:
case HitResult.SmallTickHit:
return HitResult.SmallTickHit;
case HitResult.LargeTickMiss:
case HitResult.LargeTickHit:
return HitResult.LargeTickHit;
}
return HitResult.IgnoreHit;
}
public static long GetOldStandardised(ScoreInfo score)
{
double accuracyScore =
(double)score.Statistics.Where(kvp => kvp.Key.AffectsAccuracy()).Sum(kvp => numericScoreFor(kvp.Key) * kvp.Value)
/ score.MaximumStatistics.Where(kvp => kvp.Key.AffectsAccuracy()).Sum(kvp => numericScoreFor(kvp.Key) * kvp.Value);
double comboScore = (double)score.MaxCombo / score.MaximumStatistics.Where(kvp => kvp.Key.AffectsCombo()).Sum(kvp => kvp.Value);
double bonusScore = score.Statistics.Where(kvp => kvp.Key.IsBonus()).Sum(kvp => numericScoreFor(kvp.Key) * kvp.Value);
double accuracyPortion = 0.3;
switch (score.RulesetID)
{
case 1:
accuracyPortion = 0.75;
break;
case 3:
accuracyPortion = 0.99;
break;
}
double modMultiplier = 1;
foreach (var mod in score.Mods)
modMultiplier *= mod.ScoreMultiplier;
return (long)Math.Round((1000000 * (accuracyPortion * accuracyScore + (1 - accuracyPortion) * comboScore) + bonusScore) * modMultiplier);
static int numericScoreFor(HitResult result)
{
switch (result)
{
default:
return 0;
case HitResult.SmallTickHit:
return 10;
case HitResult.LargeTickHit:
return 30;
case HitResult.Meh:
return 50;
case HitResult.Ok:
return 100;
case HitResult.Good:
return 200;
case HitResult.Great:
return 300;
case HitResult.Perfect:
return 315;
case HitResult.SmallBonus:
return 10;
case HitResult.LargeBonus:
return 50;
}
}
}
/// <summary>
/// Updates a <see cref="ScoreInfo"/> to standardised scoring.
/// This will recompite the score's <see cref="ScoreInfo.Accuracy"/> (always), <see cref="ScoreInfo.Rank"/> (always),
/// and <see cref="ScoreInfo.TotalScore"/> (if the score comes from stable).
/// The total score from stable - if any applicable - will be stored to <see cref="ScoreInfo.LegacyTotalScore"/>.
/// </summary>
/// <param name="score">The score to update.</param>
/// <param name="beatmap">The <see cref="WorkingBeatmap"/> applicable for this score.</param>
public static void UpdateFromLegacy(ScoreInfo score, WorkingBeatmap beatmap)
{
var ruleset = score.Ruleset.CreateInstance();
var scoreProcessor = ruleset.CreateScoreProcessor();
// warning: ordering is important here - both total score and ranks are dependent on accuracy!
score.Accuracy = computeAccuracy(score, scoreProcessor);
score.Rank = computeRank(score, scoreProcessor);
(score.TotalScoreWithoutMods, score.TotalScore) = convertFromLegacyTotalScore(score, ruleset, beatmap);
}
/// <summary>
/// Updates a <see cref="ScoreInfo"/> to standardised scoring.
/// This will recompute the score's <see cref="ScoreInfo.Accuracy"/> (always), <see cref="ScoreInfo.Rank"/> (always),
/// and <see cref="ScoreInfo.TotalScore"/> (if the score comes from stable).
/// The total score from stable - if any applicable - will be stored to <see cref="ScoreInfo.LegacyTotalScore"/>.
/// </summary>
/// <remarks>
/// This overload is intended for server-side flows.
/// See: https://github.com/ppy/osu-queue-score-statistics/blob/3681e92ac91c6c61922094bdbc7e92e6217dd0fc/osu.Server.Queues.ScoreStatisticsProcessor/Commands/Queue/BatchInserter.cs
/// </remarks>
/// <param name="score">The score to update.</param>
/// <param name="ruleset">The <see cref="Ruleset"/> in which the score was set.</param>
/// <param name="difficulty">The beatmap difficulty.</param>
/// <param name="attributes">The legacy scoring attributes for the beatmap which the score was set on.</param>
public static void UpdateFromLegacy(ScoreInfo score, Ruleset ruleset, LegacyBeatmapConversionDifficultyInfo difficulty, LegacyScoreAttributes attributes)
{
var scoreProcessor = ruleset.CreateScoreProcessor();
// warning: ordering is important here - both total score and ranks are dependent on accuracy!
score.Accuracy = computeAccuracy(score, scoreProcessor);
score.Rank = computeRank(score, scoreProcessor);
(score.TotalScoreWithoutMods, score.TotalScore) = convertFromLegacyTotalScore(score, ruleset, difficulty, attributes);
}
/// <summary>
/// Converts from <see cref="ScoreInfo.LegacyTotalScore"/> to the new standardised scoring of <see cref="ScoreProcessor"/>.
/// </summary>
/// <param name="score">The score to convert the total score of.</param>
/// <param name="ruleset">The <see cref="Ruleset"/> in which the score was set.</param>
/// <param name="beatmap">The <see cref="WorkingBeatmap"/> applicable for this score.</param>
/// <returns>The standardised total score.</returns>
private static (long withoutMods, long withMods) convertFromLegacyTotalScore(ScoreInfo score, Ruleset ruleset, WorkingBeatmap beatmap)
{
if (!score.IsLegacyScore)
return (score.TotalScoreWithoutMods, score.TotalScore);
if (ruleset is not ILegacyRuleset legacyRuleset)
return (score.TotalScoreWithoutMods, score.TotalScore);
var playableBeatmap = beatmap.GetPlayableBeatmap(ruleset.RulesetInfo, score.Mods);
if (playableBeatmap.HitObjects.Count == 0)
throw new InvalidOperationException("Beatmap contains no hit objects!");
ILegacyScoreSimulator sv1Simulator = legacyRuleset.CreateLegacyScoreSimulator();
LegacyScoreAttributes attributes = sv1Simulator.Simulate(beatmap, playableBeatmap);
var legacyBeatmapConversionDifficultyInfo = LegacyBeatmapConversionDifficultyInfo.FromBeatmap(beatmap.Beatmap);
var mods = score.Mods;
if (mods.Any(mod => mod is ModScoreV2))
return ((long)Math.Round(score.TotalScore / sv1Simulator.GetLegacyScoreMultiplier(mods, legacyBeatmapConversionDifficultyInfo)), score.TotalScore);
return convertFromLegacyTotalScore(score, ruleset, legacyBeatmapConversionDifficultyInfo, attributes);
}
/// <summary>
/// Converts from <see cref="ScoreInfo.LegacyTotalScore"/> to the new standardised scoring of <see cref="ScoreProcessor"/>.
/// </summary>
/// <param name="score">The score to convert the total score of.</param>
/// <param name="ruleset">The <see cref="Ruleset"/> in which the score was set.</param>
/// <param name="difficulty">The beatmap difficulty.</param>
/// <param name="attributes">The legacy scoring attributes for the beatmap which the score was set on.</param>
/// <returns>The standardised total score.</returns>
private static (long withoutMods, long withMods) convertFromLegacyTotalScore(ScoreInfo score, Ruleset ruleset, LegacyBeatmapConversionDifficultyInfo difficulty, LegacyScoreAttributes attributes)
{
if (!score.IsLegacyScore)
return (score.TotalScoreWithoutMods, score.TotalScore);
Debug.Assert(score.LegacyTotalScore != null);
if (ruleset is not ILegacyRuleset legacyRuleset)
return (score.TotalScoreWithoutMods, score.TotalScore);
double legacyModMultiplier = legacyRuleset.CreateLegacyScoreSimulator().GetLegacyScoreMultiplier(score.Mods, difficulty);
int maximumLegacyAccuracyScore = attributes.AccuracyScore;
long maximumLegacyComboScore = (long)Math.Round(attributes.ComboScore * legacyModMultiplier);
double maximumLegacyBonusRatio = attributes.BonusScoreRatio;
long maximumLegacyBonusScore = attributes.BonusScore;
double legacyAccScore = maximumLegacyAccuracyScore * score.Accuracy;
double comboProportion;
if (maximumLegacyComboScore + maximumLegacyBonusScore > 0)
{
// We can not separate the ComboScore from the BonusScore, so we keep the bonus in the ratio.
comboProportion = Math.Max((double)score.LegacyTotalScore - legacyAccScore, 0) / (maximumLegacyComboScore + maximumLegacyBonusScore);
}
else
{
// Two possible causes:
// the beatmap has no bonus objects *AND*
// either the active mods have a zero mod multiplier, in which case assume 0,
// or the *beatmap* has a zero `difficultyPeppyStars` (or just no combo-giving objects), in which case assume 1.
comboProportion = legacyModMultiplier == 0 ? 0 : 1;
}
// We assume the bonus proportion only makes up the rest of the score that exceeds maximumLegacyBaseScore.
long maximumLegacyBaseScore = maximumLegacyAccuracyScore + maximumLegacyComboScore;
double bonusProportion = Math.Max(0, ((long)score.LegacyTotalScore - maximumLegacyBaseScore) * maximumLegacyBonusRatio);
double modMultiplier = score.Mods.Select(m => m.ScoreMultiplier).Aggregate(1.0, (c, n) => c * n);
long convertedTotalScoreWithoutMods;
switch (score.Ruleset.OnlineID)
{
case 0:
if (score.MaxCombo == 0 || score.Accuracy == 0)
{
convertedTotalScoreWithoutMods = (long)Math.Round(
0
+ 500000 * Math.Pow(score.Accuracy, 5)
+ bonusProportion);
break;
}
// see similar check above.
// if there is no legacy combo score, all combo conversion operations below
// are either pointless or wildly wrong.
if (maximumLegacyComboScore + maximumLegacyBonusScore == 0)
{
convertedTotalScoreWithoutMods = (long)Math.Round(
500000 * comboProportion // as above, zero if mods result in zero multiplier, one otherwise
+ 500000 * Math.Pow(score.Accuracy, 5)
+ bonusProportion);
break;
}
// Assumptions:
// - sliders and slider ticks are uniformly distributed in the beatmap, and thus can be ignored without losing much precision.
// We thus consider a map of hit-circles only, which gives objectCount == maximumCombo.
// - the Ok/Meh hit results are uniformly spread in the score, and thus can be ignored without losing much precision.
// We simplify and consider each hit result to have the same hit value of `300 * score.Accuracy`
// (which represents the average hit value over the entire play),
// which allows us to isolate the accuracy multiplier.
// This is a very ballpark estimate of the maximum magnitude of the combo portion in score V1.
// It is derived by assuming a full combo play and summing up the contribution to combo portion from each individual object.
// Because each object's combo contribution is proportional to the current combo at the time of judgement,
// this can be roughly represented by summing / integrating f(combo) = combo.
// All mod- and beatmap-dependent multipliers and constants are not included here,
// as we will only be using the magnitude of this to compute ratios.
int maximumLegacyCombo = attributes.MaxCombo;
double maximumAchievableComboPortionInScoreV1 = Math.Pow(maximumLegacyCombo, 2);
// Similarly, estimate the maximum magnitude of the combo portion in standardised score.
// Roughly corresponds to integrating f(combo) = combo ^ COMBO_EXPONENT (omitting constants)
double maximumAchievableComboPortionInStandardisedScore = Math.Pow(maximumLegacyCombo, 1 + ScoreProcessor.COMBO_EXPONENT);
// This is - roughly - how much score, in the combo portion, the longest combo on this particular play would gain in score V1.
double comboPortionFromLongestComboInScoreV1 = Math.Pow(score.MaxCombo, 2);
// Same for standardised score.
double comboPortionFromLongestComboInStandardisedScore = Math.Pow(score.MaxCombo, 1 + ScoreProcessor.COMBO_EXPONENT);
// We estimate the combo portion of the score in score V1 terms.
// The division by accuracy is supposed to lessen the impact of accuracy on the combo portion,
// but in some edge cases it cannot sanely undo it.
// Therefore the resultant value is clamped from both sides for sanity.
// The clamp from below to `comboPortionFromLongestComboInScoreV1` targets near-FC scores wherein
// the player had bad accuracy at the end of their longest combo, which causes the division by accuracy
// to underestimate the combo portion.
// Ideally, this would be clamped from above to `maximumAchievableComboPortionInScoreV1` too,
// but in practice this appears to fail for some scores (https://github.com/ppy/osu/pull/25876#issuecomment-1862248413).
// TODO: investigate the above more closely
double comboPortionInScoreV1 = Math.Max(maximumAchievableComboPortionInScoreV1 * comboProportion / score.Accuracy, comboPortionFromLongestComboInScoreV1);
// Calculate how many times the longest combo the user has achieved in the play can repeat
// without exceeding the combo portion in score V1 as achieved by the player.
// This intentionally does not operate on object count and uses only score instead.
double maximumOccurrencesOfLongestCombo = Math.Floor(comboPortionInScoreV1 / comboPortionFromLongestComboInScoreV1);
double comboPortionFromRepeatedLongestCombosInScoreV1 = maximumOccurrencesOfLongestCombo * comboPortionFromLongestComboInScoreV1;
double remainingComboPortionInScoreV1 = comboPortionInScoreV1 - comboPortionFromRepeatedLongestCombosInScoreV1;
// `remainingComboPortionInScoreV1` is in the "score ballpark" realm, which means it's proportional to combo squared.
// To convert that back to a raw combo length, we need to take the square root...
double remainingCombo = Math.Sqrt(remainingComboPortionInScoreV1);
// ...and then based on that raw combo length, we calculate how much this last combo is worth in standardised score.
double remainingComboPortionInStandardisedScore = Math.Pow(remainingCombo, 1 + ScoreProcessor.COMBO_EXPONENT);
double scoreBasedEstimateOfComboPortionInStandardisedScore
= maximumOccurrencesOfLongestCombo * comboPortionFromLongestComboInStandardisedScore
+ remainingComboPortionInStandardisedScore;
// Compute approximate upper estimate new score for that play.
// This time, divide the remaining combo among remaining objects equally to achieve longest possible combo lengths.
remainingComboPortionInScoreV1 = comboPortionInScoreV1 - comboPortionFromLongestComboInScoreV1;
double remainingCountOfObjectsGivingCombo = maximumLegacyCombo - score.MaxCombo - score.Statistics.GetValueOrDefault(HitResult.Miss);
// Because we assumed all combos were equal, `remainingComboPortionInScoreV1`
// can be approximated by n * x^2, wherein n is the assumed number of equal combos,
// and x is the assumed length of every one of those combos.
// The remaining count of objects giving combo is, using those terms, equal to n * x.
// Therefore, dividing the two will result in x, i.e. the assumed length of the remaining combos.
double lengthOfRemainingCombos = remainingCountOfObjectsGivingCombo > 0
? remainingComboPortionInScoreV1 / remainingCountOfObjectsGivingCombo
: 0;
// In standardised scoring, each combo yields a score proportional to combo length to the power 1 + COMBO_EXPONENT.
// Using the symbols introduced above, that would be x ^ 1.5 per combo, n times (because there are n assumed equal-length combos).
// However, because `remainingCountOfObjectsGivingCombo` - using the symbols introduced above - is assumed to be equal to n * x,
// we can skip adding the 1 and just multiply by x ^ 0.5.
remainingComboPortionInStandardisedScore = remainingCountOfObjectsGivingCombo * Math.Pow(lengthOfRemainingCombos, ScoreProcessor.COMBO_EXPONENT);
double objectCountBasedEstimateOfComboPortionInStandardisedScore = comboPortionFromLongestComboInStandardisedScore + remainingComboPortionInStandardisedScore;
// Enforce some invariants on both of the estimates.
// In rare cases they can produce invalid results.
scoreBasedEstimateOfComboPortionInStandardisedScore =
Math.Clamp(scoreBasedEstimateOfComboPortionInStandardisedScore, 0, maximumAchievableComboPortionInStandardisedScore);
objectCountBasedEstimateOfComboPortionInStandardisedScore =
Math.Clamp(objectCountBasedEstimateOfComboPortionInStandardisedScore, 0, maximumAchievableComboPortionInStandardisedScore);
double lowerEstimateOfComboPortionInStandardisedScore = Math.Min(scoreBasedEstimateOfComboPortionInStandardisedScore, objectCountBasedEstimateOfComboPortionInStandardisedScore);
double upperEstimateOfComboPortionInStandardisedScore = Math.Max(scoreBasedEstimateOfComboPortionInStandardisedScore, objectCountBasedEstimateOfComboPortionInStandardisedScore);
// Approximate by combining lower and upper estimates.
// As the lower-estimate is very pessimistic, we use a 30/70 ratio
// and cap it with 1.2 times the middle-point to avoid overestimates.
double estimatedComboPortionInStandardisedScore = Math.Min(
0.3 * lowerEstimateOfComboPortionInStandardisedScore + 0.7 * upperEstimateOfComboPortionInStandardisedScore,
1.2 * (lowerEstimateOfComboPortionInStandardisedScore + upperEstimateOfComboPortionInStandardisedScore) / 2
);
double newComboScoreProportion = estimatedComboPortionInStandardisedScore / maximumAchievableComboPortionInStandardisedScore;
convertedTotalScoreWithoutMods = (long)Math.Round(
500000 * newComboScoreProportion * score.Accuracy
+ 500000 * Math.Pow(score.Accuracy, 5)
+ bonusProportion);
break;
case 1:
convertedTotalScoreWithoutMods = (long)Math.Round(
250000 * comboProportion
+ 750000 * Math.Pow(score.Accuracy, 3.6)
+ bonusProportion);
break;
case 2:
// compare logic in `CatchScoreProcessor`.
// this could technically be slightly incorrect in the case of stable scores.
// because large droplet misses are counted as full misses in stable scores,
// `score.MaximumStatistics.GetValueOrDefault(Great)` will be equal to the count of fruits *and* large droplets
// rather than just fruits (which was the intent).
// this is not fixable without introducing an extra legacy score attribute dedicated for catch,
// and this is a ballpark conversion process anyway, so attempt to trudge on.
int fruitTinyScaleDivisor = score.MaximumStatistics.GetValueOrDefault(HitResult.SmallTickHit) + score.MaximumStatistics.GetValueOrDefault(HitResult.Great);
double fruitTinyScale = fruitTinyScaleDivisor == 0
? 0
: (double)score.MaximumStatistics.GetValueOrDefault(HitResult.SmallTickHit) / fruitTinyScaleDivisor;
const int max_tiny_droplets_portion = 400000;
double comboPortion = 1000000 - max_tiny_droplets_portion + max_tiny_droplets_portion * (1 - fruitTinyScale);
double dropletsPortion = max_tiny_droplets_portion * fruitTinyScale;
double dropletsHit = score.MaximumStatistics.GetValueOrDefault(HitResult.SmallTickHit) == 0
? 0
: (double)score.Statistics.GetValueOrDefault(HitResult.SmallTickHit) / score.MaximumStatistics.GetValueOrDefault(HitResult.SmallTickHit);
convertedTotalScoreWithoutMods = (long)Math.Round(
comboPortion * estimateComboProportionForCatch(attributes.MaxCombo, score.MaxCombo, score.Statistics.GetValueOrDefault(HitResult.Miss))
+ dropletsPortion * dropletsHit
+ bonusProportion);
break;
case 3:
convertedTotalScoreWithoutMods = (long)Math.Round(
850000 * comboProportion
+ 150000 * Math.Pow(score.Accuracy, 2 + 2 * score.Accuracy)
+ bonusProportion);
break;
default:
return (score.TotalScoreWithoutMods, score.TotalScore);
}
if (convertedTotalScoreWithoutMods < 0)
throw new InvalidOperationException($"Total score conversion operation returned invalid total of {convertedTotalScoreWithoutMods}");
long convertedTotalScore = (long)Math.Round(convertedTotalScoreWithoutMods * modMultiplier);
return (convertedTotalScoreWithoutMods, convertedTotalScore);
}
/// <summary>
/// <para>
/// For catch, the general method of calculating the combo proportion used for other rulesets is generally useless.
/// This is because in stable score V1, catch has quadratic score progression,
/// while in stable score V2, score progression is logarithmic up to 200 combo and then linear.
/// </para>
/// <para>
/// This means that applying the naive rescale method to scores with lots of short combos (think 10x 100-long combos on a 1000-object map)
/// by linearly rescaling the combo portion as given by score V1 leads to horribly underestimating it.
/// Therefore this method attempts to counteract this by calculating the best case estimate for the combo proportion that takes all of the above into account.
/// </para>
/// <para>
/// The general idea is that aside from the <paramref name="scoreMaxCombo"/> which the player is known to have hit,
/// the remaining misses are evenly distributed across the rest of the objects that give combo.
/// This is therefore a worst-case estimate.
/// </para>
/// </summary>
private static double estimateComboProportionForCatch(int beatmapMaxCombo, int scoreMaxCombo, int scoreMissCount)
{
if (beatmapMaxCombo == 0)
return 1;
if (scoreMaxCombo == 0)
return 0;
if (beatmapMaxCombo == scoreMaxCombo)
return 1;
double estimatedBestCaseTotal = estimateBestCaseComboTotal(beatmapMaxCombo);
int remainingCombo = beatmapMaxCombo - (scoreMaxCombo + scoreMissCount);
double totalDroppedScore = 0;
int assumedLengthOfRemainingCombos = (int)Math.Floor((double)remainingCombo / scoreMissCount);
if (assumedLengthOfRemainingCombos > 0)
{
int assumedCombosCount = (int)Math.Floor((double)remainingCombo / assumedLengthOfRemainingCombos);
totalDroppedScore += assumedCombosCount * estimateDroppedComboScoreAfterMiss(assumedLengthOfRemainingCombos);
remainingCombo -= assumedCombosCount * assumedLengthOfRemainingCombos;
if (remainingCombo > 0)
totalDroppedScore += estimateDroppedComboScoreAfterMiss(remainingCombo);
}
else
{
// there are so many misses that attempting to evenly divide remaining combo results in 0 length per combo,
// i.e. all remaining judgements are combo breaks.
// in that case, presume every single remaining object is a miss and did not give any combo score.
totalDroppedScore = estimatedBestCaseTotal - estimateBestCaseComboTotal(scoreMaxCombo);
}
return estimatedBestCaseTotal == 0
? 1
: 1 - Math.Clamp(totalDroppedScore / estimatedBestCaseTotal, 0, 1);
double estimateBestCaseComboTotal(int maxCombo)
{
if (maxCombo == 0)
return 1;
double estimatedTotal = 0.5 * Math.Min(maxCombo, 2);
if (maxCombo <= 2)
return estimatedTotal;
// int_2^x log_4(t) dt
estimatedTotal += (Math.Min(maxCombo, 200) * (Math.Log(Math.Min(maxCombo, 200)) - 1) + 2 - Math.Log(4)) / Math.Log(4);
if (maxCombo <= 200)
return estimatedTotal;
estimatedTotal += (maxCombo - 200) * Math.Log(200) / Math.Log(4);
return estimatedTotal;
}
double estimateDroppedComboScoreAfterMiss(int lengthOfComboAfterMiss)
{
if (lengthOfComboAfterMiss >= 200)
lengthOfComboAfterMiss = 200;
// int_0^x (log_4(200) - log_4(t)) dt
// note that this is an pessimistic estimate, i.e. it may subtract too much if the miss happened before reaching 200 combo
return lengthOfComboAfterMiss * (1 + Math.Log(200) - Math.Log(lengthOfComboAfterMiss)) / Math.Log(4);
}
}
private static double computeAccuracy(ScoreInfo scoreInfo, ScoreProcessor scoreProcessor)
{
int baseScore = scoreInfo.Statistics.Where(kvp => kvp.Key.AffectsAccuracy())
.Sum(kvp => kvp.Value * scoreProcessor.GetBaseScoreForResult(kvp.Key));
int maxBaseScore = scoreInfo.MaximumStatistics.Where(kvp => kvp.Key.AffectsAccuracy())
.Sum(kvp => kvp.Value * scoreProcessor.GetBaseScoreForResult(kvp.Key));
return maxBaseScore == 0 ? 1 : baseScore / (double)maxBaseScore;
}
public static ScoreRank ComputeRank(ScoreInfo scoreInfo) => computeRank(scoreInfo, scoreInfo.Ruleset.CreateInstance().CreateScoreProcessor());
private static ScoreRank computeRank(ScoreInfo scoreInfo, ScoreProcessor scoreProcessor)
{
var rank = scoreProcessor.RankFromScore(scoreInfo.Accuracy, scoreInfo.Statistics);
foreach (var mod in scoreInfo.Mods.OfType<IApplicableToScoreProcessor>())
rank = mod.AdjustRank(rank, scoreInfo.Accuracy);
return rank;
}
/// <summary>
/// Used to populate the <paramref name="score"/> model using data parsed from its corresponding replay file.
/// </summary>
/// <param name="score">The score to run population from replay for.</param>
/// <param name="files">A <see cref="RealmFileStore"/> instance to use for fetching replay.</param>
/// <param name="populationFunc">
/// Delegate describing the population to execute.
/// The delegate's argument is a <see cref="SerializationReader"/> instance which permits to read data from the replay stream.
/// </param>
public static void PopulateFromReplay(this ScoreInfo score, RealmFileStore files, Action<SerializationReader> populationFunc)
{
string? replayFilename = score.Files.FirstOrDefault(f => f.Filename.EndsWith(@".osr", StringComparison.InvariantCultureIgnoreCase))?.File.GetStoragePath();
if (replayFilename == null)
return;
try
{
using (var stream = files.Store.GetStream(replayFilename))
{
if (stream == null)
return;
using (SerializationReader sr = new SerializationReader(stream))
populationFunc.Invoke(sr);
}
}
catch (Exception e)
{
Logger.Error(e, $"Failed to read replay {replayFilename} during score migration", LoggingTarget.Database);
}
}
private class FakeHit : HitObject
{
private readonly Judgement judgement;
public override Judgement CreateJudgement() => judgement;
public FakeHit(Judgement judgement)
{
this.judgement = judgement;
}
}
private class FakeJudgement : Judgement
{
public override HitResult MaxResult { get; }
public FakeJudgement(HitResult maxResult)
{
MaxResult = maxResult;
}
}
}
}