示例#1
0
        public void ValidateProbabilityLn(
            [Values(1.0, 1.0, 2.0, 2.0, 2.0, 2.0)] double lambda,
            [Values(1.0, 1.0, 1.0, 1.0, 2.0, 2.0)] double nu,
            [Values(1, 2, 1, 2, 1, 3)] int x,
            [Values(-1.0, -1.69314718055995, -1.30685281944005, -1.30685281944005, -0.754324797564617, -2.95154937490084)] double pln)
        {
            var d = new ConwayMaxwellPoisson(lambda, nu);

            AssertHelpers.AlmostEqual(pln, d.ProbabilityLn(x), 13);
        }
示例#2
0
        public override void ExecuteExample()
        {
            // <a href="http://en.wikipedia.org/wiki/Binomial_distribution">Binomial distribution</a>
            MathDisplay.WriteLine("<b>Binomial distribution</b>");
            // 1. Initialize the new instance of the Binomial distribution class with parameters P = 0.2, N = 20
            var binomial = new Binomial(0.2, 20);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Binomial distribution class with parameters P = {0}, N = {1}", binomial.P, binomial.N);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", binomial);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", binomial.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", binomial.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", binomial.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", binomial.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", binomial.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", binomial.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", binomial.Mean.ToString(" #0.00000;-#0.00000"));

            // Median
            MathDisplay.WriteLine(@"{0} - Median", binomial.Median.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", binomial.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", binomial.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", binomial.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", binomial.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Binomial distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Binomial distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(binomial.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Bernoulli_distribution">Bernoulli distribution</a>
            MathDisplay.WriteLine("<b>Bernoulli distribution</b>");
            // 1. Initialize the new instance of the Bernoulli distribution class with parameter P = 0.2
            var bernoulli = new Bernoulli(0.2);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Bernoulli distribution class with parameter P = {0}", bernoulli.P);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", bernoulli);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", bernoulli.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", bernoulli.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", bernoulli.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", bernoulli.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", bernoulli.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", bernoulli.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", bernoulli.Mean.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", bernoulli.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", bernoulli.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", bernoulli.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", bernoulli.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Bernoulli distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Bernoulli distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(bernoulli.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Categorical_distribution">Categorical distribution</a>
            MathDisplay.WriteLine("<b>Categorical distribution</b>");
            // 1. Initialize the new instance of the Categorical distribution class with parameters P = (0.1, 0.2, 0.25, 0.45)
            var binomialC = new Categorical(new[] { 0.1, 0.2, 0.25, 0.45 });

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Categorical distribution class with parameters P = (0.1, 0.2, 0.25, 0.45)");
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", binomialC);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", binomialC.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", binomialC.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", binomialC.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", binomialC.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", binomialC.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", binomialC.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", binomialC.Mean.ToString(" #0.00000;-#0.00000"));

            // Median
            MathDisplay.WriteLine(@"{0} - Median", binomialC.Median.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", binomialC.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", binomialC.StdDev.ToString(" #0.00000;-#0.00000"));

            // 3. Generate 10 samples of the Categorical distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Categorical distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(binomialC.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Conway%E2%80%93Maxwell%E2%80%93Poisson_distribution">ConwayMaxwellPoisson distribution</a>
            MathDisplay.WriteLine("<b>Conway Maxwell Poisson distribution</b>");
            // 1. Initialize the new instance of the ConwayMaxwellPoisson distribution class with parameters Lambda = 2, Nu = 1
            var conwayMaxwellPoisson = new ConwayMaxwellPoisson(2, 1);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the ConwayMaxwellPoisson distribution class with parameters Lambda = {0}, Nu = {1}", conwayMaxwellPoisson.Lambda, conwayMaxwellPoisson.Nu);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", conwayMaxwellPoisson);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", conwayMaxwellPoisson.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", conwayMaxwellPoisson.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", conwayMaxwellPoisson.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", conwayMaxwellPoisson.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", conwayMaxwellPoisson.Mean.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", conwayMaxwellPoisson.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", conwayMaxwellPoisson.StdDev.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the ConwayMaxwellPoisson distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the ConwayMaxwellPoisson distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(conwayMaxwellPoisson.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Discrete_uniform">DiscreteUniform distribution</a>
            MathDisplay.WriteLine("<b>Discrete Uniform distribution</b>");
            // 1. Initialize the new instance of the DiscreteUniform distribution class with parameters LowerBound = 2, UpperBound = 10
            var discreteUniform = new DiscreteUniform(2, 10);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the DiscreteUniform distribution class with parameters LowerBound = {0}, UpperBound = {1}", discreteUniform.LowerBound, discreteUniform.UpperBound);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", discreteUniform);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", discreteUniform.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", discreteUniform.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", discreteUniform.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", discreteUniform.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", discreteUniform.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", discreteUniform.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", discreteUniform.Mean.ToString(" #0.00000;-#0.00000"));

            // Median
            MathDisplay.WriteLine(@"{0} - Median", discreteUniform.Median.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", discreteUniform.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", discreteUniform.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", discreteUniform.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", discreteUniform.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the DiscreteUniform distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the DiscreteUniform distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(discreteUniform.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Geometric_distribution">Geometric distribution</a>
            MathDisplay.WriteLine("<b>Geometric distribution</b>");
            // 1. Initialize the new instance of the Geometric distribution class with parameter P = 0.2
            var geometric = new Geometric(0.2);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Geometric distribution class with parameter P = {0}", geometric.P);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", geometric);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", geometric.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", geometric.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", geometric.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", geometric.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", geometric.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", geometric.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", geometric.Mean.ToString(" #0.00000;-#0.00000"));

            // Median
            MathDisplay.WriteLine(@"{0} - Median", geometric.Median.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", geometric.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", geometric.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", geometric.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", geometric.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Geometric distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Geometric distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(geometric.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Hypergeometric_distribution">Hypergeometric distribution</a>
            MathDisplay.WriteLine("<b>Hypergeometric distribution</b>");
            // 1. Initialize the new instance of the Hypergeometric distribution class with parameters PopulationSize = 10, M = 2, N = 8
            var hypergeometric = new Hypergeometric(30, 15, 10);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Hypergeometric distribution class with parameters Population = {0}, Success = {1}, Draws = {2}", hypergeometric.Population, hypergeometric.Success, hypergeometric.Draws);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", hypergeometric);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", hypergeometric.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", hypergeometric.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", hypergeometric.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", hypergeometric.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", hypergeometric.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", hypergeometric.Mean.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", hypergeometric.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", hypergeometric.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", hypergeometric.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", hypergeometric.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Hypergeometric distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Hypergeometric distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(hypergeometric.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Negative_binomial">NegativeBinomial distribution</a>
            MathDisplay.WriteLine("<b>Negative Binomial distribution</b>");
            // 1. Initialize the new instance of the NegativeBinomial distribution class with parameters P = 0.2, R = 20
            var negativeBinomial = new NegativeBinomial(20, 0.2);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the NegativeBinomial distribution class with parameters P = {0}, N = {1}", negativeBinomial.P, negativeBinomial.R);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", negativeBinomial);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", negativeBinomial.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", negativeBinomial.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", negativeBinomial.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", negativeBinomial.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", negativeBinomial.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", negativeBinomial.Mean.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", negativeBinomial.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", negativeBinomial.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", negativeBinomial.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", negativeBinomial.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the NegativeBinomial distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the NegativeBinomial distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(negativeBinomial.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Poisson_distribution">Poisson distribution</a>
            MathDisplay.WriteLine("<b>Poisson distribution</b>");
            // 1. Initialize the new instance of the Poisson distribution class with parameter Lambda = 1
            var poisson = new Poisson(1);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Poisson distribution class with parameter Lambda = {0}", poisson.Lambda);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", poisson);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", poisson.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", poisson.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", poisson.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", poisson.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", poisson.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", poisson.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", poisson.Mean.ToString(" #0.00000;-#0.00000"));

            // Median
            MathDisplay.WriteLine(@"{0} - Median", poisson.Median.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", poisson.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", poisson.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", poisson.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", poisson.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Poisson distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Poisson distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(poisson.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();


            // <a href="http://en.wikipedia.org/wiki/Zipf_distribution">Zipf distribution</a>
            MathDisplay.WriteLine("<b>Zipf distribution</b>");
            // 1. Initialize the new instance of the Zipf distribution class with parameters S = 5, N = 10
            var zipf = new Zipf(5, 10);

            MathDisplay.WriteLine(@"1. Initialize the new instance of the Zipf distribution class with parameters S = {0}, N = {1}", zipf.S, zipf.N);
            MathDisplay.WriteLine();

            // 2. Distributuion properties:
            MathDisplay.WriteLine(@"2. {0} distributuion properties:", zipf);

            // Cumulative distribution function
            MathDisplay.WriteLine(@"{0} - Сumulative distribution at location '3'", zipf.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            MathDisplay.WriteLine(@"{0} - Probability mass at location '3'", zipf.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            MathDisplay.WriteLine(@"{0} - Log probability mass at location '3'", zipf.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Entropy
            MathDisplay.WriteLine(@"{0} - Entropy", zipf.Entropy.ToString(" #0.00000;-#0.00000"));

            // Largest element in the domain
            MathDisplay.WriteLine(@"{0} - Largest element in the domain", zipf.Maximum.ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            MathDisplay.WriteLine(@"{0} - Smallest element in the domain", zipf.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            MathDisplay.WriteLine(@"{0} - Mean", zipf.Mean.ToString(" #0.00000;-#0.00000"));

            // Mode
            MathDisplay.WriteLine(@"{0} - Mode", zipf.Mode.ToString(" #0.00000;-#0.00000"));

            // Variance
            MathDisplay.WriteLine(@"{0} - Variance", zipf.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            MathDisplay.WriteLine(@"{0} - Standard deviation", zipf.StdDev.ToString(" #0.00000;-#0.00000"));

            // Skewness
            MathDisplay.WriteLine(@"{0} - Skewness", zipf.Skewness.ToString(" #0.00000;-#0.00000"));
            MathDisplay.WriteLine();

            // 3. Generate 10 samples of the Zipf distribution
            MathDisplay.WriteLine(@"3. Generate 10 samples of the Zipf distribution");
            for (var i = 0; i < 10; i++)
            {
                MathDisplay.Write(zipf.Sample().ToString("N05") + @" ");
            }
            MathDisplay.FlushBuffer();
            MathDisplay.WriteLine();
            MathDisplay.WriteLine();
        }
        public void ValidateProbabilityLn(double lambda, double nu, int x, double pln)
        {
            var d = new ConwayMaxwellPoisson(lambda, nu);

            AssertHelpers.AlmostEqual(pln, d.ProbabilityLn(x), 13);
        }
 public void ValidateProbabilityLn(double lambda, double nu, int x, double pln)
 {
     var d = new ConwayMaxwellPoisson(lambda, nu);
     AssertHelpers.AlmostEqualRelative(pln, d.ProbabilityLn(x), 12);
 }
        /// <summary>
        /// Run example
        /// </summary>
        /// <a href="http://en.wikipedia.org/wiki/Conway%E2%80%93Maxwell%E2%80%93Poisson_distribution">ConwayMaxwellPoisson distribution</a>
        public void Run()
        {
            // 1. Initialize the new instance of the ConwayMaxwellPoisson distribution class with parameters Lambda = 2, Nu = 1
            var binomial = new ConwayMaxwellPoisson(2, 1);

            Console.WriteLine(@"1. Initialize the new instance of the ConwayMaxwellPoisson distribution class with parameters Lambda = {0}, Nu = {1}", binomial.Lambda, binomial.Nu);
            Console.WriteLine();

            // 2. Distributuion properties:
            Console.WriteLine(@"2. {0} distributuion properties:", binomial);

            // Cumulative distribution function
            Console.WriteLine(@"{0} - Сumulative distribution at location '3'", binomial.CumulativeDistribution(3).ToString(" #0.00000;-#0.00000"));

            // Probability density
            Console.WriteLine(@"{0} - Probability mass at location '3'", binomial.Probability(3).ToString(" #0.00000;-#0.00000"));

            // Log probability density
            Console.WriteLine(@"{0} - Log probability mass at location '3'", binomial.ProbabilityLn(3).ToString(" #0.00000;-#0.00000"));

            // Smallest element in the domain
            Console.WriteLine(@"{0} - Smallest element in the domain", binomial.Minimum.ToString(" #0.00000;-#0.00000"));

            // Mean
            Console.WriteLine(@"{0} - Mean", binomial.Mean.ToString(" #0.00000;-#0.00000"));

            // Variance
            Console.WriteLine(@"{0} - Variance", binomial.Variance.ToString(" #0.00000;-#0.00000"));

            // Standard deviation
            Console.WriteLine(@"{0} - Standard deviation", binomial.StdDev.ToString(" #0.00000;-#0.00000"));
            Console.WriteLine();

            // 3. Generate 10 samples of the ConwayMaxwellPoisson distribution
            Console.WriteLine(@"3. Generate 10 samples of the ConwayMaxwellPoisson distribution");
            for (var i = 0; i < 10; i++)
            {
                Console.Write(binomial.Sample().ToString("N05") + @" ");
            }

            Console.WriteLine();
            Console.WriteLine();

            // 4. Generate 100000 samples of the ConwayMaxwellPoisson(4, 1) distribution and display histogram
            Console.WriteLine(@"4. Generate 100000 samples of the ConwayMaxwellPoisson(4, 1) distribution and display histogram");
            var data = new double[100000];

            for (var i = 0; i < data.Length; i++)
            {
                data[i] = binomial.Sample();
            }

            ConsoleHelper.DisplayHistogram(data);
            Console.WriteLine();

            // 5. Generate 100000 samples of the ConwayMaxwellPoisson(2, 1) distribution and display histogram
            Console.WriteLine(@"5. Generate 100000 samples of the ConwayMaxwellPoisson(2, 1) distribution and display histogram");
            binomial.Lambda = 2;
            for (var i = 0; i < data.Length; i++)
            {
                data[i] = binomial.Sample();
            }

            ConsoleHelper.DisplayHistogram(data);
            Console.WriteLine();

            // 6. Generate 100000 samples of the ConwayMaxwellPoisson(5, 2) distribution and display histogram
            Console.WriteLine(@"6. Generate 100000 samples of the ConwayMaxwellPoisson(5, 2) distribution and display histogram");
            binomial.Lambda = 5;
            binomial.Nu     = 2;
            for (var i = 0; i < data.Length; i++)
            {
                data[i] = binomial.Sample();
            }

            ConsoleHelper.DisplayHistogram(data);
        }
 public void ValidateProbabilityLn(
     [Values(1.0, 1.0, 2.0, 2.0, 2.0, 2.0)] double lambda, 
     [Values(1.0, 1.0, 1.0, 1.0, 2.0, 2.0)] double nu, 
     [Values(1, 2, 1, 2, 1, 3)] int x, 
     [Values(-1.0, -1.69314718055995, -1.30685281944005, -1.30685281944005, -0.754324797564617, -2.95154937490084)] double pln)
 {
     var d = new ConwayMaxwellPoisson(lambda, nu);
     AssertHelpers.AlmostEqual(pln, d.ProbabilityLn(x), 13);
 }