codes.tune.evaluate_tuning

codes.tune.evaluate_tuning#

Functions

compute_hypervolume_over_time(study[, ...])

Compute the 2D hypervolume curve over time while excluding the last N trials from the Pareto calculation (by completion time).

compute_pareto_plot_data(study, suffix[, ...])

evaluate_tuning(study_prefix[, top_n, ...])

For all surrogate studies named '<study_prefix>_<surrogate>', plot the top_n test-loss trajectories.

format_surrogate_title(name)

get_best_trials(study, top_n)

Return trial numbers of best 'top_n' trials.

hypervolume_2d(pareto_points, reference)

load_loss_history(model_path)

Load loss histories from a saved .pth.

main()

pareto_front(points)

pareto_front_mask(points)

parse_args()

plot_losses(loss_histories, epochs, labels)

Plot multiple loss trajectories on a log scale.

save_individual_pareto_plot(data, out_dir)

save_pareto_front_grid(datasets, out_dir)

save_relative_hv_combined(datasets, out_dir)

save_relative_hv_grid(datasets, out_dir)

summarize_pareto_tradeoff(data)