Discussion of article "Population optimization algorithms: Saplings Sowing and Growing up (SSG)" - page 13
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
unfortunately, or maybe fortunately, the bot does not generate new information, but the information that is publicly available is often distorted. i think this is due to the mechanisms of information interpolation. it lies, makes up non-existent abbreviations of algorithms, even makes up the names of authors on the fly and the dates of the algorithm)))). you should be very careful with such information.
As an assistant in text editing, stylistic edits and getting reference when writing articles - yes, an indispensable assistant.
Thank you. I indirectly find local ones through forced interruption of optimisation when a large number of cores are involved. Roughly speaking, there are 20 agents in the Tester, I interrupt optimisation after 2000 passes.
Hereis an example of sets obtained with such interruption. If not interrupted, the pictures on the link would show all 20 sets the same. But with interruption you can see different behaviour of sets, among which there may well be those that pass OOS.
If we find 20 local extrema (I suggested the method of gradual ejection), then displaying these extrema on such a picture will give the most objective visual assessment of the TS.
For self-education, what is the dependence of complexity on measurement?
I confess - I don't know. I only know that it grows non-linearly fast.
Aleksey Nikolavev appeared here, maybe he knows the exact answer to this question. I forgot which way to call a forum user.
Exact knowledge is hardly possible here at all, only some estimates.
1) Growth of the number of seats in relation to the number of extrema. Let's assume a smooth case (a discontinuous variant can always be approximated by a smooth one with some accuracy). The extremum is at points with degeneracy of the gradient and is determined by the signs of the eigennumbers of the hessian. Let the dimension N and (let us assume for simplicity) each of the signs of the eigenvalues of the hessian is determined by a random choice with equal probabilities 0.5, then the probability that all signs are the same (so it is an extremum) is 2/(2^N)=2^(1-N). For the two-dimensional case it will be equal to 0.5 (50%), which is good and quite visible in the pictures - the number of saddles is approximately equal to the number of extrema. In the 10-dimensional case, the extrema will already be less than 0.2%.
2) In fact, any algorithm for finding extrema creates a dynamical system, which tend to be more and more chaotic with increasing dimensions. You may recall that the Mandelbrot set arises in a dynamical system, which arises when iteratively searching for the root of a quadratic function in the two-dimensional case) A large number of seats only contributes to even more chaotic system.
Exact knowledge is hardly even possible here, only some kind of estimate.
1) Growth of the number of saddles with respect to the number of extrema. Let us assume a smooth case (a discontinuous variant can always be approximated by a smooth one with some accuracy). The extremum is at points with degeneracy of the gradient and is determined by the signs of the eigennumbers of the hessian. Let the dimension N and (let us assume for simplicity) each of the signs of the eigenvalues of the hessian is determined by a random choice with equal probabilities 0.5, then the probability that all signs are the same (so it is an extremum) is 2/(2^N)=2^(1-N). For the two-dimensional case it will be equal to 0.5 (50%), which is good and quite visible in the pictures - the number of saddles is approximately equal to the number of extrema. In the 10-dimensional case the extrema will be already less than 0.2%.
2) In fact, any algorithm for finding extrema creates a dynamical system, which tend to be more and more chaotic as the dimensions grow. You may recall that the Mandelbrot set arises in a dynamical system, which arises when iteratively searching for the root of a quadratic function in the two-dimensional case) A large number of seats only contributes to even more chaotic system.
Quite pessimistic calculation for multidimensional variants is obtained.
The calculation for multivariate variants is rather pessimistic.
In general, yes. That is why usually in multidimensional cases the task of a complete study of the surface device of the loss function is not set. Neither is the task of searching for the global extremum. In fact, one is limited to just finding good enough points. Well, maybe except for cases where it is possible to construct a loss function with good properties, as in MO, for example.
I have a problem with MT5, which does not allow me to access the folder: open Data Folder, in addition to that, I do not have the full elements on the platform MT 5 customizable, I mean by that I do not have the full tools such as rectangle, square or and triangle, etc.. So please, if anyone knows anything about solving these problems, please give me a shout.
you wont get any help by posting on an thread that is unrelated to your post. Create your own thread. And as this is a technical forum, you need to provide more information than you have given including the first 3 lines of the journal when you open mt5 that has your system details.
However, without any more information, it sounds to me like a corrupted installation. try uninstalling mt5, and delete any folders that remain including folders such as "config" that is often not uninstalled correctly. Then try installing it again.