EGTtools: Toolbox for Evolutionary Game Theory

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

Full Changelog: https://github.com/Socrats/EGTTools/compare/v0.1.9...v0.1.10