CMA-ES/pycma r4.6.0

on GitHub

• Improve the return value when constraints are passed to fmin2 similar to
  fmin_con2.
• Option 'verb_disp_overwrite' remains inactive if 'verb_disp' is True
  or integer 1.
• Provide set_popsize method to change the population size during a run.
• Allow dynamic setting of learning rates by assigning values or callbacks
  c*_dynamic of a CMAEvolutionStrategy instance. The callback allows to
  use the last used value for the new computation.
• Change the computation for lower std bounds of integer variables to improve
  the semantics of integer_std_lower_bound, setting
  integer_std_lower_bound_distance=1/3 and making the binary flag
  obsolete.
• Methods to check whether a genotype solution is in the pre-bounds
• CMAOptions._stationary_sphere can be a function that returns a new mean
• Polish various fitness function implementations
• Various smaller bug and corner-case fixes including sigma-redistribution,
  feed_for_resume, rotation seed 0, numpy quirks, mainly by @antonsoo

Full Changelog: r4.5.0...r4.6.0