Tags

* project:
  * add [codecov.io](https://codecove.io) to produce coverage reports in travis
  * use Visual Studio 2019 image in appveyor
  * [boost/ut](https://github.com/boost-experimental/ut): use boost.UT v1.1.6
  * [fmtlib/fmt](https://fmt.dev/latest/index.html): bump to version v6.1.1

* optimizer:
  * use a evolutionary algorithm to init/reinit solver. In optimization mode, Baryonyx stores all solutions found during search. Two algorithms are used to initialize the X vector:
    * A crossover/mutation using two solutions is used to re-initialize the solver.
    * An improve mode using a dynamical kappa.
  * Quadratic objective function: Experimental support.
  * Z-coefficient: Enable the Z-coefficient solver.
  * random-solver: This solver is enabled with the `--random` command line option
  and in the `solver_parameters::solver_type`. It provides a pure random solver without any preference and Lagrangian relaxation. It is provided only for benchmark.
  * add Lagrangian multipliers sign change detection.

* c++:
  * solvers: replace `fill` and `copy` with `fill_n` and `copy_n` equivalent.
  * solvers: replace `std::vector<T>` with a new `bit_array` to improve cache  efficiency
  * main: fix signed/unsigned warning

* solver
  * negative coefficient solvers: fix R access
  * add pi signed change detection
  * improve efficiency of random-solver
  * add reset function to reinit solver
  * solver: move normalized cost in thread part

* optimizer:
  * replace time limit detection algorithm
  * add pi signed change detection to build a new constraint remaining list

* benchmark:
  * fix sort functor behaviour

* main:
  * fix time-limit parameter assignation
  * missing or malformed parameters equal error
  * fix infinity norm string
  * add a `--debug` mode.

* CMake:
  * reduce warning level for msvc
  * use the BUILD_SHARED_LIBS CMake option to build share library. Default, a
     static library is build to enable all internal test that use private API.

Bump C++ dependency to C++17 standard.

* benchmark: add a second output `csv` file with the rank for each solvers and
  models to show best solver.

* Optimize all solver algorithms to improve performance: rewrite and
  generalize.

* Improve the `best-cycle` algorithm: is no best solution was found for three
  complete optimization, solver restart to `bastert-cycle`.

* Fix the initialization cycle behaviour.

* Fix the random normalization behaviour.

* Remove useless header (sparse-vector etc), dead code.

v0.3.4

* Add a benchmark mode to permit to run benchmark on a set of problems described in a `csv` files and make easy comparison.
* Improve colour rendering in log system. Replace specific Linux/Unix source code about colour in (pseudo)terminal with the `fmt::text_style` and `fmt::terminal_color` enumeration.
* Add warning message for unknown parameters

* core: add init policy cycle:
  * `bastert` (default): use cost values.
  * `random`: use random value.
  * `best`: use best solution found if available (`bastert` otherwise).
  * `bastert-cycle`: start with bastert policy then start a cycle between random, best and bastert.
  * `random-cycle`: start with random policy then start a cycle between best, bastert and random.
  * `best-cycle`: start with best policy then start a cycle between bastert, random and best.

* core: add lagrangian runtime constraint order.
  * `lagrangian-decr`: sort violated constraints according to the Lagrangian multiplier values in decremental order.
  * `lagrangian-incr`: sort violated constraints according to the Lagrangian multiplier values in incremental order

baryonyx v0.3.3 - 2018-10-04

main:

* output: better render output float number

solver/optimizer:

* Fix the preference matrix assignment for both equalities and inequalities
with -1 and 1 coefficients, we replace the preference matrix clean up algorithm
(invert the variable values for negative coefficients) with a version that
revert only the added or subtracted value during the preferences update
process.
* Fix core dump if no solution are found

preprocessor:

* update the preprocessing solver option: Replace the old preprocessing option
with new one but without description (p1, p2, p3 and p4). Options are `none` to
use the raw problem (or lp file) order for constraints and variables. `memory`,
the default, to use the raw problem (or lp file) order for constraints but sort
the variables to improve the memory cache efficiency.

v0.3.2

v0.3.1

v0.3.0.0

Version used to bench Baryonyx for the ROADEF 2018 conference

v0.2.3.0

v0.2.2.0

v0.2.1.0

v0.2.0.0

v0.1.0.0