Changelog

Source: NEWS.md

QGA 1.1

• Introduced parallelization (optional, parameter parallel, default FALSE)

• Added the quantum interference crossover operator (parameters interference_flag, default FALSE, and interference_rate, default 0.1)

• Added interference_type = "amplitude" (with strength interference_lambda): fitness-weighted interference that sums the qubit amplitudes of the population, constructively for individuals above the average fitness and destructively for those below

• Added the vignette “Avoiding local optima”, illustrating the parameters that counteract premature convergence and how to use them

• Added three options against premature convergence (all off by default): the H-epsilon gate (epsilon), which keeps the probability of each bit value in [epsilon, 1 - epsilon]; a partial restart towards equal probabilities after restart_iters generations without improvement (restart_strength); personal attractors with local and global migration (attractor_type = "personal", migration_group_size, migration_local_period, migration_global_period)

• Fixed a bug in the rotation gate: when an individual was at least as good as the best one, the case x = 1, b = 0 was never handled and the case x = 0, b = 1 was rotated twice in opposite directions

• Added two parameters to the QGA function:

  • stop_limit (value to stop the iterations if the fitness is higher than a given value)
  • stop_iters (number of iterations without improvement of fitness before stopping)