Measurement Metrics for Multi-Objective Optimizations

Подписаться 5 тыс.

Просмотров 5 тыс.

50% 1

Measurement Metrics for Multi-Objective Optimizations
To design an optimization or define suitable stop criteria for optimization runs measurements for the quality of the search are mandatory. When it comes to multi-objective optimization (MOO) the amount of possible criteria is much higher due to a growing space of possibilities.
In this video we present you the two most common metrics for MOO and explain you how they work. Furthermore, you get some advice how to use them when the real pareto frontier is unknown.
Python package: pymoo.org
LinkedIn: / fabianrang
GitLab: gitlab.com/you...
-------------------------------------------------------------------------------
Data Science to go: paretos.com

Опубликовано:

9 сен 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 7

@niklaschief3774 3 года назад

Video fast so stabil wie die Frisur ❤️

@MrRugbyferdinand 3 года назад

Very nice and intuitive explanation!👍 Could you maybe specifiy why it is important that your points on the pareto front are well distributed and how that should help improving the performance of the model? Thanks in advance!

@mauip.7742 3 года назад

finally learnt something about the Measurement Metrics ! Ehre geht raus

@mohamedantar6865 Год назад

Thanks you are awesome Plz explain the c-metric GD, IGD

@anata.one.1967 Год назад

Can I average the hyper volume over a number of runs to get an statistically relevant performance metrics?

@Bencurlis 3 года назад

Great video! Imagine you give a final population size as input to a heuristic, let's say 50, and it produces less non-dominated solutions, let's say 30. How do you choose on which population you use the measure (whole population or non-dominated subset)? Obviously for some measure it should give the same value, e.g. hypervolume I guess, but surely that won't be the case in general. And how do you deal with measures that require to know the Pareto front like GD and IGD? Also, how many points should be taken on the Pareto front? Same size as the non-dominated population, or same size as the input population size, maybe much more?