New Post: Negative Samples from Geometric Distribution

Christoph,

I meant to also mention an alternative way of displaying histograms in test methods (I find it useful to include output in tests to easily observe the various distribution configurations).

I basically just extended ConsoleHelper with a variety of additional methods to make the output compatible with the MSTest output restrictions (no random console cursor access). The main method for a histogram that can be displayed in test output is shown below.

(I have added a variety of other methods that embellish the output with basic statistics and whatnot, but those have nothing to do with the console output restrictions solved here).

The output for a histogram simply needs to be rotated 90 degrees to make it work reasonably well in tests:

        public static void DisplayHistogramRotate90(IEnumerable<double> data
            , int numBuckets = 20, int maxBarLength = 80)
        {
            var fmt = "{0, 6} : {1, -" + maxBarLength + "} : {2}";
            var blockSymbol = Convert.ToChar(9608);
            var hist = new Histogram(data, numBuckets);

            var maxBucketCount = 0.0;
            for (var i = 0; i < hist.BucketCount; i++)
            {
                if (hist[i].Count > maxBucketCount)
                    maxBucketCount = hist[i].Count;
            }

            Console.WriteLine("Histogram:");
            for (var i = 0; i < hist.BucketCount; i++)
            {
                var fraction = hist[i].Count / maxBucketCount;
                var dots = (int)(maxBarLength * fraction);
                Console.WriteLine(fmt, i, new string(blockSymbol, dots), hist[i]);
            }
        }

Also, I added a method to generalize the display for any particular distribution (when there is a chance of throwing a "NotImplementedException" when accessing certain properties):

        public static void DisplayDiscreteDistribution(IDiscreteDistribution dist, int numHistSamples = 100000, int numHistBuckets = 20)
        {
            const string fmt = "{0, 20} = {1, 20}";
            var divider = new string('*', 120);
            Console.WriteLine(divider);
            Console.WriteLine("{0}:\n", dist);

            try
            {
                Console.WriteLine(fmt, "CDF (location 3)", dist.CumulativeDistribution(3).ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Prob (location 3)", dist.Probability(3).ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "ProbLn (location 3)", dist.ProbabilityLn(3).ToString("f5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Entropy", dist.Entropy.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "MaxValue", dist.Maximum.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "MinValue", dist.Minimum.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Mean", dist.Mean.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Mode", dist.Mode.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Variance", dist.Variance.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "StdDev", dist.StdDev.ToString("F5"));
            }
            catch { }
            try
            {
                Console.WriteLine(fmt, "Skewness", dist.Skewness.ToString("F5"));
            }
            catch { }
            Console.WriteLine();

            Console.Write("Samples(10): ");
            for (var i = 0; i < 10; i++)
            {
                Console.Write(dist.Sample().ToString("N5") + " ");
            }
            Console.WriteLine();
            Console.WriteLine();

            //// 4. Generate 100000 samples of the distribution and display histogram
            var data = new double[numHistSamples];
            for (var i = 0; i < data.Length; i++)
            {
                data[i] = dist.Sample();
            }

            DisplayHistogramRotate90(data, numHistBuckets);
            Console.WriteLine(divider);
        }

Obviously, there is another similar method for IContinuousDistribution types.

So my tests then can then look something like the following (with no assertions in this case):

       [TestMethod]
        public void GeometricVariationsTest()
        {
             var dist = new Geometric(p: 0.01, randomSource: new MersenneTwister());
            ConsoleHelper.DisplayDiscreteDistribution(dist, 100000, 40);

            dist.P = 0.2;
            ConsoleHelper.DisplayDiscreteDistribution(dist, 100000, 40);

            dist.P = 0.5;
            ConsoleHelper.DisplayDiscreteDistribution(dist, 100000, 40);

            dist.P = 0.8;
            ConsoleHelper.DisplayDiscreteDistribution(dist, 100000, 40);
        }

Hopefully, someone else might find this a useful way of experimenting with the distributions, without needing to create repetitive console application code.
;-)