Use a different way to initialize the PRNG creation for the .NET
framework as described on this SO answer:
https://stackoverflow.com/a/38644970
Add System.Security.Cryptography.Algorithms NuGet on netstandard1.6.
* BetaRegularize Fix Issue 491
Changed the number of iterations for the loop from 140 to 100,000. This
brought the result from
MathNet.Numerics.SpecialFunctions.BetaRegularized(6985172036240.918,
5029323866093.4609, 0.581395348837209); to 0.5101 which is inline with
the results in R(0.5) and Mathematica(0.495).
* I suspect they fail because the algorithm doesn't work well
for those test cases, not that there is an implementation error.
But this suspicion has not been verified.
* Changed file and class names from TestClassName to
ClassNameTests to better conform with other parts of Math.Net
* Added some organizational abstractions to simplify adding
test cases.
* Added test cases based on several functions from
Testing Unconstrained Optimization Software
Jorge J. Moré, Burton S. Garbow, Kenneth E. Hillstrom
ACM Transactions on Mathematical Software, Vol 7, No. 1, March 1981, Pages 17-41.
* At this point, most test functions are low-dimensional,
well-scaled, unimodal functions.
* There are test failures for Conjugate Gradient and Newton
unconstrained minimizers. Currently I believe these to be
failures of the algorithm, rather than failures of
implementation, but it hasn't been investigated thoroughly.
* Previous versions of the class did not allow subclasses to
initialize the value/gradient/hessian fields once and reuse the
memory for subsequent evaluations.
* This commit exposes raw value/gradient/hessian fields to
subclasses