Source of failure was MATH-1355: "EigenDecomposition" was assuming that a diagonal matrix could be made non-diagonal.
Added a utility method that creates a dense matrix from an array specifying its diagonal elements.
New sampling API for multivariate distributions (similar to changes performed for MATH-1158).
Unit test file renamed in accordance to the class being tested.
One failing test "@Ignore"d (see comments on the bug-tracking system).
Adaptor from new to old API.
Class is deprecated: it is temporarily provided to allow testing of the new RNGs (in user applications) with minimal changes.
New package "o.a.c.m.rng" contains RNG core functionality (uniform distribution).
Utilities to interface with stress test suites.
Userguide: benchmarks and stress test suites reports (MATH-1327).
Removed deprecated sampling API (and constructors).
In this pass, the integer distributions have been updated.
Class "RandomDataGenerator" has been reverted to use the old RNG implementation (a wrapper is created whenever it must be passed to a distribution's "createSampler" method).
Previously integration would crash when two event detectors triggered events
within the root finder's tolerance of each other. Now all events are handled
in order and integration proceeds.
Fixes MATH-1342
Using the new sampler API.
Deprecate old API (including the now obsolete constructors).
Apart from the main focus of the issue (distribution classes), this change also had an impact on:
* some of test classes that use random numbers (where the tolerance may have its value updated when it is highly sensitive on the RNG seed)
* statistical inference tests that use a distribution but not the RNG (cf. MATH-1124) which now deprecated code passed to its contructor.
* classes in package "o.a.c.m.random" that depend on some of the distributions