Discussing the article: "Role of random number generator quality in the efficiency of optimization algorithms" - page 11
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Here you 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 off afterwards.
You're showing ignorance of the topic, sort out the difference between runs and restarts first. My post is about a rule of thumb for choosing the number of FF runs for optimisation algorithms, i.e. how many runs of a fitness function should be run to ensure acceptable convergence of the optimisation algorithms. In my tests I use 10000 thousand runs of the fitness function, which corresponds to 10^4. In this article we considered the influence of the GSC quality on the results of optimisation at these 10000 runs of the fitness function as a part of the optimisation algorithms, the influence of the GSC quality was not revealed, as it is stated in the conclusions to the article.
And for the comparison of DSTs among themselves, a test on the uniformity of the distribution of random numbers, with a much larger number of runs of DSTs than 10^4.
2024.03.18 20:54:33.459 Standard, 100000000 throws, 10000 boxes
I.e., 100000000 = 10^8!
Please don't post more in the comments of my articles.
You're showing ignorance of the topic, understand the difference between runs and restarts first. My post is about a rule of thumb for choosing the number of FF runs for optimisation algorithms, i.e. how many runs of a fitness function should be run to ensure acceptable convergence of optimisation algorithms. In my tests I use 10000 thousand runs of the fitness function, which corresponds to 10^4. In this article we have considered the influence of the quality of the HSC on the results of optimisation at these 10000 runs of the fitness function as part of the optimisation algorithms, the influence of the quality of the HSC has not been revealed, as it is stated in the conclusions to the article.
And for comparison of DSTs with each other, a test for uniformity of random number distribution was carried out with a much larger number of DST runs than 10^4.
2024.03.18 20:54:33.459 Standard, 100000000 throws, 10000 boxes
I.e., 100000000 = 10^8!
Please don't post more in the comments of my articles.
To compare the same optimisation alg. with different DSTs, you need to make about the same number of its restarts, not 5. The more the better. Then it will be clear which DST is better/worse/how it affects on average.
5 is the number of tests. Optimisation algorithms give a spread of results greater than the apparent difference from using different DSTs.
First you claimed, in deleted posts earlier, that there shouldn't be an effect of DST, and now you claim that there is an effect, but it is not revealed in tests? You are confusing your statements.
Once again I repeat, conduct your own tests and either refute my conclusions or confirm them. I have provided all the tools for the tests. Verbal statements are not interesting to anyone here, my articles are for practical application, not for the sake of theory in a vacuum.
Please don't write more in comments to my articles.
5 is the number of tests. Optimisation algorithms give a variation in the results more than the visible difference from the application of different DSTs.
First you claimed, in deleted posts earlier, that there should not be any influence of DGS, and now you claim that there is an influence, but it is not revealed in the tests? You are confusing your statements.
Once again I repeat, conduct your own tests and either refute my conclusions or confirm them. I have provided all the tools for the tests.
Please do not write more in comments to my articles.
Even such a frontal option would work for me. But I don't understand how to define the area of poking in a multidimensional space?
Any thoughts, how to determine the area of the found global peak by the calculated values (let them be 10 000 pieces) of FF? To force -DBL_MAX in this area on the next iteration of FF.
The Expectation Maximisation (in general) and Gaussian Mixture Model (in particular) algorithm class is ideal for this. It will select all hills as separate clusters, after which you can zero any of them.
Unfortunately, this is not available in MQL5, even in ALGLIB. For now, you can pull it from python. If you have a desire, the materials on this site will also work.
The Expectation Maximisation (in general) and Gaussian Mixture Model (in particular) class of algorithms is ideal for this. It will select all hills as separate clusters, after which any of them can be zeroed.
Unfortunately, this is not available in MQL5, even in ALGLIB. For now, you can pull it from python. If you have a desire, the materials on this site will also work.
The question is what to do next with the set of these sets with "hill tops". Earlier we had one global maximum as a solution of the optimisation algorithm, let's say now we have 50 of them. But they don't come close to solving the stability problem.
Well, we don't know why Saber is looking for peaks (maybe he will tell us), in his problem statement maybe this is the way to find robust solutions.
In the diagram I gave earlier, what element in the diagram do you think affects finding a robust solution (robust Result)?
In the diagram I gave earlier, which element in the diagram do you think affects finding a robust solution (robust Result)?
In my view, there is none.
The way I see it, there isn't one here.