/// <summary>
/// Initializes a new instance of the <see cref="ToyExperiment" /> class.
/// </summary>
/// <param name="model">The model.</param>
/// <param name="game">The game.</param>
/// <param name="priors">The priors.</param>
public ToyExperiment(IModel model, Game game, Marginals priors)
{
Model = model;
Game = game;
Priors = priors;
Posteriors = Model.Train(Game, Game.Players, Priors).Posteriors;
}
/// <summary>
/// Demonstration of Dynamics.
/// </summary>
/// <param name="trueBeta">The true beta.</param>
/// <param name="beta">The beta.</param>
/// <param name="stepSize">Size of the step.</param>
public void DynamicsDemo(double trueBeta, double beta, double stepSize)
{
var randomNames = new RandomNameGenerator(0, int.MaxValue, true);
var players = new HashSet <string>(randomNames.Take(100));
string improver = players.First();
const double Mu = 125;
const double Sigma = 10;
var skillDistribution = new Gaussian(Mu, Sigma * Sigma);
const double Gamma = 0.4;
const double SkillStart = 110;
const double SkillMax = 140;
var inputs1 = new Inputs <TwoPlayerGame> {
Mu = Mu, Sigma = Sigma, Beta = beta, Gamma = 0
};
var inputs2 = new Inputs <TwoPlayerGame> {
Mu = Mu, Sigma = Sigma, Beta = beta, Gamma = Gamma
};
double truePerformanceVariance = trueBeta * trueBeta;
// number of games
const int NumberOfGames = 25000;
var skill1 = Gaussian.PointMass(SkillStart);
var random = new Random(0);
var initialSkills = players.ToDictionary(ia => ia, ia => Gaussian.PointMass(skillDistribution.Sample()));
initialSkills[improver] = skill1;
var trueSkills = new Dictionary <string, List <double> > {
{ improver, new List <double>() }
};
for (int i = 0; i < NumberOfGames; i++)
{
// Pick two players to play game
var gamePlayers = players.OrderBy(ia => random.Next()).Take(2).ToList();
// update improver's skill
var truth = new Marginals {
Skills = new Dictionary <string, Gaussian>(initialSkills)
};
truth.Skills[improver] = skill1;
// Generate game according to truth
var game = TwoPlayerVaryingSkills.Sample(truth, gamePlayers, truePerformanceVariance);
inputs1.Games.Add(game);
inputs2.Games.Add(game);
if (gamePlayers.Contains(improver))
{
trueSkills[improver].Add(skill1.Point);
// Update skill
if (trueSkills[improver].Count > 100 && (trueSkills[improver].Count % 100) >= 90 && skill1.Point < SkillMax)
{
skill1 = Gaussian.PointMass(skill1.Point + stepSize);
}
}
}
var experiment1 = new OnlineExperiment(ia => new TwoPlayerVaryingSkills(ia, ShowFactorGraph), inputs1.TrueSkillParameters)
{
Players = players,
Priors = inputs1.TrueSkillPriors,
Name = "Fixed skill model"
};
var experiment2 = new OnlineExperiment(ia => new TwoPlayerVaryingSkills(ia, ShowFactorGraph), inputs2.TrueSkillParameters)
{
Players = players,
Priors = inputs2.TrueSkillPriors,
Name = "Varying skill model"
};
AnnounceExperiment(experiment1.Name);
experiment1.Run(inputs1.Games, inputs1.Games.Count, false);
AnnounceExperiment(experiment2.Name);
experiment2.Run(inputs2.Games, inputs2.Games.Count, false);
var experiments1 = new OnlineExperimentComparison <TwoPlayerGame>(experiment1)
{
Truth = trueSkills
};
var experiments2 = new OnlineExperimentComparison <TwoPlayerGame>(experiment2)
{
Truth = trueSkills
};
var initialSkillsTable =
initialSkills.OrderByDescending(ia => ia.Key)
.ToDictionaryForTable <KeyValuePair <string, Gaussian>, string, double, double>(
ia => ia.Key,
new Dictionary <string, Func <KeyValuePair <string, Gaussian>, double> > {
{ "Skill", g => g.Value.Point }
},
false,
null,
"Player");
var collection = new OnlineExperimentComparison <TwoPlayerGame>(experiment1, experiment2)
{
Truth = trueSkills
};
var dynamicsDemo = new Dictionary <string, object>
{
{ "InitialSkills", initialSkillsTable },
{ "SampledInputs", inputs1 },
{
"Trajectories",
new[]
{
experiments1.PlayerTrajectoriesWithSigmaAndTruth,
experiments2.PlayerTrajectoriesWithSigmaAndTruth,
collection.PlayerTrajectoriesWithSigmaAndTruth
}
}
};
outputter.Out(dynamicsDemo,
Contents.S5AllowingTheSkillsToVary.NumberedName,
"DynamicsDemo");
}
