Egttools

- removed support for win32 and manylinux_i686 for Python > 3.7

Added

- added new controls to `draw_stationary_distribution`
- added `enhancement_factor` parameter to `CRDGame`. This parameter serves as a multiplying factor for the payoff of
each player in case the target is achieved. If `enhancement_factor = 1` the `CRDGame` behaves as usual.
For `enhancement_factor > 1`, we are incentivizing strategies that reach the target.
- added `MemoryOneStrategy` to `egttools.games.NormalForm.TwoAction` strategies.
- added `CommonPoolResourceDilemma` game - However it has still not been extensively tested!!
- added `ninja` as a requirement for the build.
- added `TimeBasedCRDStrategy` to `egttools.games.CRD` strategies. These strategies make contributions to the Public
Good in function of the round of the game.
- added `sdist` to build.
- added labels to the lines plot by `plot_gradients` so that several lines can be plotted.
- added more unit testing, but this still needs a lot of improvement.
- added missing libraries on C++ code.

Fixed

- fixed issue with environmental variables not working on MacOS builds in Github Actions
- fixed gradient of selection estimation algorithm
- fixed issue with multiple `libomp.dylib` copies loading
- fixed issue with the citation of current version of egttools
- fixed link to anaconda documentation

Changed

- now openmp is linked dynamically to avoid conflicts on MacOS
- increased minimum required cmake version to `3.18`
- changed int to `int_fast64_t` to support larger number of generations

Added

- added NetworkSync evolver
- added more support for network simulations
- added Cache to accelerate `PairwiseComparison`
- added calculation of the averaged gradient of selection
- added OneShotCRDNetworkGame

[0.1.12.patch1] - 7-07-2023

Fixed

- Fixed issue with link to Simplex2D class
- Match scatter output to matplotlib API changes [by cvanelteren]
- Fixed issue with compiling `Distributions.cpp` without Boost

Changed

- Removed support for `Python 3.7`
- Dropped support for `Win32` architectures
- Removed `MACOSX_DEPLOYMENT_TARGET` constraint
- Updated egttools citation
- Skip `multivariate_hypergeometric` tests when compiling without Boost
- Upgraded cibuildwheel to v2.14.0
- Set minimum Boost version to 1.70.0

Added

- Added basic unittests for plotting [by cvanelteren]

Fixed

- Fixed stability plot and minor syntax issues.
- Fixed issue with fitness calculations of strategies with 0 counts in the population.
- Fixed LNK2005 error on windows.
- Fixed version naming convention to adhere to PEP440.

Changed

- Changed name of `PairwiseMoran` class to `PairwiseComparisonNumerical`.
- Changed function and class names that used "moran" to "pairwise_comparison_rule".
- Removed signatures of overloaded methods in docstrings.
- Moved all headers for pybind11 to .hpp files to avoid issues on windows.
- Updated pybind11 version to 2.10.
- Removed specialization of `binomialCoeff` to avoid issues on Windows.
- Refactored the binding code from C++ to Python so that now we use different files for defining each Python submodule.
This makes the code much more clear. These files can be found in `cpp/src/pybind11_files`.
- Now the GitHub CI builds download and install Boost C++ library, as it is required to build egttools.
- Dropped support for manylinux_i686. In the future only 64 bit architectures will be supported.

Added

- Added support for multiprecision types with boost::multiprecision.
- Added replicator equation for n-player games (implementation in c++).
- Added `replicator_equation` implementation in C++.
- Added a new C++ implementation of the analytical stochastic dynamics named `PairwiseComparison`.
- Added `Matrix2PlayerGameHolder` and `MatrixNPlayerGameHolder` classes, so that a game can be defined using a payoff
matrix.
- Added several tests (coverage now is ~50%).
- Added tutorials to the documentation.

[0.1.11.patch2] - 16-11-2022

Fixed

- Fixed error in `calculate_full_transition_matrix`. The previous implementation tried to calculate the fitness for
strategies with 0 counts in the population. This would lead to an error when calculating the probability of group
forming. Now, we calculate directly the probability of a strategy with 0 counts increasing by 1 individual in the
population, i.e., prob = (n_decreasing_strategy / population_size) * (mu / (1 - nb_strategies)).

[0.1.11.patch1] - 10-11-2022

Fixed

- Fixed issue with changing mutation rate in StochDynamics. When the mutation rate was updated
using `evolver.mu = new_value`, the variable `evolver.not_mu` was not udpated. This variable was used
by `calculate_full_stationary_distribution` and it created an error. Now `not_mu` is calculated inside
of `calculate_full_stationary_distribution`.
- Fixed issue with `binomialCoeff` implemented in c++. In some cases, the intermediary calculations were overflowing
an `uint64_t` type when calling it from `multivariateHypergeometricPDF`. Now, we no longer calculate the exact value
in `multivariateHypergeometricPDF`, instead use double types. This will be improved in the new version by
using `boost/multiprecision` `uint128_t` types.

Fixed

- fixed errors in docstrings examples
- fixed missing headers
- fixed error in `full_fitness_difference_group` and `calculate_fulll_transition_matrix`. There was a missing
multiplying factor in the probability of transition due to mutation. The probability of selecting the strategy to dies
must also be taken into account. There was also an issue when instantiating the `multi_hypergeometric_distribution`
class from scipy. It does not copy the array containing the counts of each strategy. Now we create a copy before
passing the state vector to avoid the issue.
- fixed issue with `AbstractNPlayerGame`. For N-player games it was not a good idea to calculate the fitness in Python
as this part of the class becomes a bottleneck, as it will be called more often then in the 2-player case (because
there are more states - so less likely the fitness will be stored in cache). For this reason we now implemented this
abstract class in C++ and the fitness calculation is done in this language. Everything else remains the same, and it
should be equally easy to create new games.
- fixed issues of missing initialization of internal parameters of some `egttools.games.NormalForm.TwoAction`
strategies. Internal parameters should be initialized when `time == 0`, i.e., at the beginning of each game.

Changed

- changed `egttools` to `src-layout`. This should help fix issues with tests and make the overall structure of the
library cleaner.
- moved C++ code to the `cpp` folder. This way the code is more organized.
- Bump pypa/cibuildwheel from 2.8.1 to 2.11.1
- Bump pypa/cibuildwheel from 2.11.1 to 2.11.2

Fixed

- fixed issue with the comparison of different types in `PairwiseMoran`
- fixed issue with colorbar so that now the axis is passed to maptlolib colorbar, this way it will be plotted correctly
when drawing multiple subplots
- fixed issue with hardcoded x-axis in `plot_gradients`
- fixed error on `calculate_full_transition_matrix`. The error happened when calculating the transition probability:
- a) although the literature is a bit confusing on this, the original paper by Traulsen 2006 says that for the
transition we consider that the strategies to reproduce and die are pricked simultaneously (so both with
probability 1/Z).
- b) the is more accumulated numerical error when doing probability of transitioning from A to B P(A, B) = fermi(
-beta, B - A) than when doing fermi(beta, A - B). This is probably specific to Python and Numpy, but must be taken
into account in the future.
- c) The schur decomposition (egttools.utils.calculate_stationary_distribution_non_hermitian) works better in this
case (although still has a slight numerical error) and should be used for full transition matrices).
- normalized transition probabilities to use the definition in Traulsen et al. 2006
- now we assume that both death and birth individuals are selected simultaneously with probability n_i/Z, where n_i
is the number of individuals of that strategy in the population

Changed

- updated installation instructions
- updated to PEP 621 syntax
- updated setup.py since now scikit-build supports VS2019
- updated `draw_stationary_distribution` to make the display of labels optional
- changed stability calculation for the replicator dynamics to use the Jacobian matrix
- updated `plot_gradients` to check for all possible types of roots (stable, unstable and saddle)
- removed stability checks for the stochastic dynamics
- if T+ is too small, phi will be approximated to infinity and the fixation probability will be approximated to 0.
- This may not be correct, since if p_minus is also very small or equal to p_plus, the outcome would be different.
So it might change in a future version
- updated default language for documentation to `en`
- updated docstrings
- changed colorbar default label to gradient of selection
- droped pin to Sphinx <= 4.5.0
- updated variable name in AbstractGame
- changed name of variable in `calculate_fitness` method of Abstract game

Added

- added input parameter checks for `run` and `evolve` methods of `PairwiseMoran`
- created a method to calculate roots
- created a method to check the stability of the replicator dynamics through the Jacobian matrix
- added a check for the limit case in which the only non-negative eigenvalue is close to atol
- added new notebook with examples of use
- added an extra check when calculating fixation probabilities
- added Python 3.10 binary - except for Windows and manylinuxi686 as non numpy or scipy builds yet available
- added new CRD strategy
- added extra tolerance controls in `check_replicator_stability_pairwise_games`
- added new abstract game classes to simplify game implementation
- added `NPlayerStagHunt` game

[0.1.9-patch6] - 16-02-2022

Fixed

- Fixed error on version formatting, it should be `0.1.9.dev6` instead of `0.1.9.patch6`.

[0.1.9-patch5] - 16-02-2022

Fixed

- Fixed wrong version tag on git.

[0.1.9-patch4] - 16-02-2022

Fixed

- There was a problem with setting a `geometric_distribution` in C++ as a private variable for OpenMP which caused some
errors when estimating stationary distributions. This was fixed by setting it as a shared variable.

Added

- Binder links to run examples and updated notebooks

[0.1.9-patch3] - 03-02-2022

Added

- Added `gitter` chat and `binder` launch.

Fixed

- Added missing `seaborn` dependency. This dependency is only needed to be able to automatically generate colorblind
colors to plot the invasion diagram, so it might be dropped in the future. But, for the moments, it provides the
easiest way to do this.

Changed

- Updated docs and notebooks to use the latest `egttools` API.

[0.1.9-patch2] - 26-01-2022

Fixed

- This release fixes an issue with a modulo operation that was causing an index error when calculating stability for
plotting dynamics on a simplex.

[0.1.9-patch1] - 19-01-2022

Fixed

- This release fixes an issue with CITATION.cff which prevented zenodo from publishing a new doi.