Discussing the article: "Role of random number generator quality in the efficiency of optimization algorithms" - page 10

New comment
 
Andrey Dik #:

fractality

The distance between vertices must not be below a given threshold.

 
I don't think it is possible to find strictly all locales, 50 means 50. But multimodal AOs should solve such problems well enough and find all significant locales (within the constraints specified in the problem), I mean, of course, by optimisation algorithm for a reasonable number of FF runs.

 
Andrey Dik #:
I don't think it is possible to find strictly all locales, 50 means 50. But multimodal AOs should solve such problems well enough and find all significant locales (within the constraints specified in the problem), I mean, of course, by optimisation algorithm for a reasonable number of FF runs.

If it finds less - no big deal.

 

I have met such an empirical formula for the required number of FF launches:

Fcount=D*10^4

where:

Fcount - number of FF runs

D - number of measurements, i.e. of optimised parameters

Well, I don't know, for what very weak AO this formula is used, i.e. for one parameter 10000 FF runs!

The formula is linear, rough, but can be applied in quite. I suppose even D*10^3 is normal.

 
You're pathetic 😀😀😀😀
Repeating after me by deleting my comments.
 
Dick has a monopoly on commenting on articles? What about "discussion" and pluralism of opinion?
 
I demand to return my comment, in which it was written that for an adequate comparison of HCS algorithms requires a billion restarts, a multiple of the number of parameters!

And that the testing in the article is given incorrectly and misleads the reader.
 
Maxim Dmitrievsky #:
I demand to return my comment, in which it was written that for an adequate comparison of HCS algorithms requires a billion restarts, a multiple of the number of parameters!

And that the testing in the article is given incorrectly and misleads the reader.

It is not enough what you can say or claim.
Do a correct test from your point of view and either refute or confirm my conclusions in the article.
Flooding, insults and emoji abuse are not welcome here. Constructive suggestions and experimentally validated criticism are welcome.
 
Andrey Dik #:

There is little you can say or assert.
Do a correct test from your point of view and either refute or confirm my conclusions in the article.
Flooding, insults and emoji abuse are not welcome here. Constructive suggestions and experimentally validated criticism are welcome.
Run a valid test to compare gsx. Ask chatgpt if you don't know how.
 
Andrey Dik #:

I have come across such an empirical formula for the required number of FF runs:

Fcount=D*10^4

where:

Fcount - number of FF launches

D - number of measurements, i.e. optimised parameters

Well, I don't know, for which very weak AO this formula is used, i.e. for one parameter 10000 FF runs!

The formula is linear, rough, but can be applied in quite. I suppose even D*10^3 is normal.

Here you yourself answer your own question, how many minimum restarts should be approximately to compare different DSTs. This is exactly what I wrote about and you brushed it up later.
1...3456789101112131415
New comment