pomma89/troschuetz.random

Fully managed library providing various random number generators and distributions.

troschuetz.random's Issues

The behavior of TRandom.Next(int, int) is different from the built-in's.

As https://msdn.microsoft.com/en-us/library/2dx6wyd4(v=vs.110).aspx mentioned:

Remarks

The Next(Int32, Int32) overload returns random integers that range from minValue to maxValue – 1. However, if maxValue equals minValue, the method returns minValue.

It should not raise an exception when (minValue == maxValue).

First Int is always 1

In the following scenario, results will always be { 0, 100000, 0, 0 }.

            var results = new int[4];
            for (int x = 0; x < 100000; x++)
            {
                var rnd = new TRandom();
                results[rnd.Next(4)]++;
            }

In the next scenario, results will be something like { 24971, 124881, 25068, 25080 }.

            var results = new int[4];
            for (int x = 0; x < 100000; x++)
            {
                var rnd = new TRandom();
                results[rnd.Next(4)]++;
                results[rnd.Next(4)]++;
            }

So the first roll is always 1. The second is sufficiently random.

Why are the uint and int functions opposite in terms of default bounds?

Looking at IGenerator, I see that Next() produces an int between 0, and int.MaxValue exclusive, and NextInclusiveMaxValue() is between 0 and int.MaxValue inclusive. Looking at the uint functions, I see the opposite: NextUInt() produces a value between 0 and uint.MaxValue inclusive, whereas NextUIntExclusiveMaxValue() produces a value between 0 and uint.MaxValue exclusive.

Why does NextUInt() default to an inclusive bound, whereas Next() defaults to an exclusive bound? At first glance it seems a little unintuitive -- I would expect that both would default to the same bounds behavior, if there is no reason to do otherwise.

MT19937 not as in Wikipedia?

I have implemented MT19937 in C# according to the code in wikipedia and wanted to verify results with your implementation, but they were wrong.
I narrowed it to the line
https://github.com/pomma89/Troschuetz.Random/blob/fe9423531c158ee01ad1d16515f186fb5d7ac9ca/src/Troschuetz.Random/Generators/MT19937Generator.cs#L353
It reduces random number range from 32 to 31 bits.

In wikipedia case it is

     y := y xor (y >> l)
     return lowest w bits of (y)

Notice it is "L" (18) not 1.
that I translated as

            y = y ^ (y >> l);
            return 0xffffffffU & y;

Notice that your results are also different from the online test

MIT License?!

I would like to use this library in a project that is under MIT. Do you think it would be feasible to change the license of Troschuetz.Random to MIT?