Discussing the article: "Population optimization algorithms: Changing shape, shifting probability distributions and testing on Smart Cephalopod (SC)"

New comment
 

Check out the new article: Population optimization algorithms: Changing shape, shifting probability distributions and testing on Smart Cephalopod (SC).

The article examines the impact of changing the shape of probability distributions on the performance of optimization algorithms. We will conduct experiments using the Smart Cephalopod (SC) test algorithm to evaluate the efficiency of various probability distributions in the context of optimization problems.

Working on this article and specific class methods for generating random numbers with the necessary distributions, convenient for use in building optimization algorithms, led to the understanding that the Rastrigin function has several serious shortcomings that were not obvious at the time of choosing this test function, so I decided not to use it. The good old Rastrigin will be replaced by the Peaks function (a more complete justification will be provided in the next article).

Peaks1

"Smart Cephalopod" in action

Author: Andrey Dik

 
While reading an article, the ln-number appears abruptly and then is mentioned frequently. What is it?
 
fxsaber #:
While reading an article, the ln-number appears abruptly and then is mentioned frequently. What is it?

in, input, input, input.
There is some number on a numeric line, relative to which we need to generate a random number with a given distribution, and within given bounds.
It is often necessary to do this in optimisation algorithms.

"For example, we have this situation with a numerical line bounded on the left by "min" and on the right by "max" and lying between them by "In", we can visualise it as follows:

min|-----|-----|-----|-----|-----|-----|In|--|--|--|--|--|--|max"


 
Andrey Dik #:
in, input, input, input.
There's a perception problem with that letter. It's like it's written in logarithm.
 
fxsaber #:
There's a perception problem with this letter. It's like a logarithm.
Maybe it is - a typo, I'll check.
 
Andrey Dik #:
That may be the case - typo, I'll check.

No, it's not a typo, it's just the font is such that the capital "I" reads like an "l", corrected it to "i".

Thanks for the comment.

 

Распределение Леви является примером распределения с неограниченными моментами. Оно описывает случайные величины с тяжелыми хвостами, что означает, что вероятность появления очень больших значений велика. В распределении Леви моменты могут быть бесконечными или не существовать, что делает его особенным и отличающимся от распределений с ограниченными моментами, таких как нормальное распределение.

The heavy tails of the Levy distribution and its unbounded moments make it useful for modelling phenomena that may have extreme values or high variability.

At the beginning of the article, definitions of the main terms were given. It would be good to add this one as well.

 
fxsaber #:

The definitions of the main terms were given at the beginning of the article. It would be good to add this one as well.

I will, thank you.
 

If I understood correctly, almost all articles showed different search strategies. At the same time there was no play with distributions.

This article shows that the results can strongly depend on the chosen distribution function + bias.


Judging by the numbers, Levy ripped everyone apart. It turns out that for each search strategy, you have to make several modifications and see the effect on the overall ranking.


The author is awesome! Thanks!


I commented out the switch to distributions.

//    revision = true;

and got a better result than the uniform distribution.

 

In the MT5 GA, all inputs are hard-coded and the distribution seems to be simply uniform.


Thanks to this series of articles it became clear that results can vary greatly not only from the search strategy, but also from its input parameter values. Plus setting distributions.

How to find the optimal one for your task is not quite clear. Because you need to optimise what is being optimised.

 
fxsaber uniform.
Please show how to measure the quality of the MT5 GA on test functions.
12
New comment