Treeskill mit Teams in Infer.net

Post a reply

Smilies
:) :( :oops: :chelo: :roll: :wink: :muza: :sorry: :angel: :read: *x) :clever:
View more smilies
BBCode is ON
[img] is ON
[flash] is OFF
[url] is ON
Smilies are ON
Topic review
   

Expand view Topic review: Treeskill mit Teams in Infer.net

by Guest » 18 Feb 2025, 12:32

Ich versuche, ein Trueskill -Modell mit jeweils zwei Teams und 5 Spielern mit Infer.net zu erstellen. Wenn Sie jedoch die Fähigkeiten schließen, werden die Mittel der Verteilung viel zu groß oder klein. Obwohl ich hier ausgefehnte Spieler und Übereinstimmungen hier verwende, erhalte ich ähnliche Ergebnisse mit Daten aus der realen Welt. < /P>

Code: Select all

using Microsoft.ML.Probabilistic.Distributions;
using Microsoft.ML.Probabilistic.Factors;
using Microsoft.ML.Probabilistic.Models;
using Microsoft.ML.Probabilistic.Collections;
using Microsoft.Data.Analysis;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Numerics;
using System.Text;
using System.Threading.Tasks;
using System.Collections;
using System.Security.Cryptography;
using System.IO;
using System.Security.Policy;
using System.Reflection;

namespace ConsoleApp3
{
internal class infernetDemo5
{
static void Main(string[] args)
{

int[][] winnerData = {
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 5 },
new[] { 5, 6, 7, 8, 9 },
new[] { 5, 6, 7, 8, 9 },
};
int[][] loserData = {
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 5, 6, 7, 8, 9 },
new[] { 6, 7, 8, 9, 4 },
new[] { 0, 1, 2, 3, 4 },
new[] { 0, 1, 2, 3, 4 },
};

var nGames = 32;
var nPlayers = 10;
var mu = 25.0;
var sigma = 8.333;
var beta = 4.1667;

var teamSize = new Range(5).Named("TeamSize");
var games = new Range(nGames).Named("Game");
var players = new Range(nPlayers).Named("Player");

var playerSkills = Variable.Array(players).Named("Skill");

var winner_lineups = Variable.Array(Variable.Array(teamSize), games).Named("t1Lineups");
var loser_lineups = Variable.Array(Variable.Array(teamSize), games).Named("t2Lineups");

var w_performances = Variable.Array(Variable.Array(teamSize), games);
var l_performances = Variable.Array(Variable.Array(teamSize), games);

using (Variable.ForEach(players))
{
playerSkills[players] = Variable.GaussianFromMeanAndPrecision(mu,  1 / (sigma* sigma));

}

using (var game = Variable.ForEach(games))
{
var gameIndex = game.Index;
using (var n = Variable.ForEach(teamSize))
{
var playerIndex = n.Index;
w_performances[gameIndex][playerIndex] = Variable.GaussianFromMeanAndPrecision(playerSkills[winner_lineups[gameIndex][playerIndex]], 1 / (beta* beta)).Named("w_player_performance");
l_performances[gameIndex][playerIndex] = Variable.GaussianFromMeanAndPrecision(playerSkills[loser_lineups[gameIndex][playerIndex]], 1 / (beta* beta)).Named("l_player_perfomance");
}

}

var w_performance = Variable.Sum(w_performances[games]);
var l_performance = Variable.Sum(l_performances[games]);
Variable.ConstrainTrue(w_performance > l_performance);

winner_lineups.ObservedValue = winnerData;
loser_lineups.ObservedValue = loserData;

var inferenceEngine = new InferenceEngine();
inferenceEngine.NumberOfIterations = 10;
var inferredSkills = inferenceEngine.Infer(playerSkills);

foreach(var playerSkill in inferredSkills)
{
Console.WriteLine(playerSkill);
}

}

}
}
< /code>
Ausgabe: < /p>
Compiling model...done.
Iterating:
.........| 10
Gaussian(25, 69,44)
Gaussian(25, 69,44)
Gaussian(25, 69,44)
Gaussian(25, 69,44)
Gaussian(-6,574e+04, 26,68)
Gaussian(-4,68e+04, 14,8)
Gaussian(-4,889e+04, 10,78)
Gaussian(-4,889e+04, 10,78)
Gaussian(-4,889e+04, 10,78)
Gaussian(-4,889e+04, 10,78)
Verwenden von trueskillthroughtime https://github.com/glandfried/tueskillthroughtime.py mit denselben Beispieldaten und -werten, die demselben Modell entsprechen sollten, ergibt die folgenden Ergebnisse:

Code: Select all

N(mu=26.223, sigma=3.618)
N(mu=26.223, sigma=3.618)
N(mu=26.223, sigma=3.618)
N(mu=26.223, sigma=3.618)
N(mu=16.821, sigma=3.618)
N(mu=33.179, sigma=3.618)
N(mu=23.777, sigma=3.618)
N(mu=23.777, sigma=3.618)
N(mu=23.777, sigma=3.618)
N(mu=23.777, sigma=3.618)
Ich vermute, dass das Zusammenfassen der Spieler des Teams das Problem ist, wenn die Einschränkung der einzelnen Leistungen gemäß den Ergebnissen angemessene Fähigkeitenschätzungen gibt.

Top