The following changes were made to improve MIPROv2 logging:
Ensuring that both mini-batching and full eval programs (along with their scores) are saved at each trial in trial_logs. This fixes a previous issue where the full eval score was being overwritten.
Returning evaluated programs and their scores in best_program.candidate_programs (all fully evaluated programs) and best_program.mb_candidate_programs (only minibatch evaluated programs), for consistency with Random Search Few-Shot
The following changes were made to improve MIPROv2 logging:
trial_logs
. This fixes a previous issue where the full eval score was being overwritten.best_program.candidate_programs
(all fully evaluated programs) andbest_program.mb_candidate_programs
(only minibatch evaluated programs), for consistency with Random Search Few-Shot