Algorithm Optimisation Championship. - page 12
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
A final clarification of the problem.
It was said that it would be necessary to find 100 and 500 maxima of FF, as well as - global maximum.
I understood it as follows: I need to find 500 "small" peaks, 100 "big" peaks, and one "absolute" peak.
Bottom line: you need to find 601 - but the peak value of the FF.
Right?
A final clarification of the problem.
It was said that it would be necessary to find 100 and 500 maxima of FF, as well as - global maximum.
I understood it as follows: I need to find 500 "small" peaks, 100 "big" peaks, and one "absolute" peak.
Bottom line: you need to find 601 - but the peak value of the FF.
Right?
A final clarification of the problem.
It was said that it would be necessary to find 100 and 500 maxima of FF, as well as - global maximum.
I understood it as follows: I need to find 500 "small" peaks, 100 "big" peaks, and one "absolute" peak.
Bottom line: you need to find 601 - but the peak value of the FF.
Right?
:)
Where did you read that? I'm very interested, really.
ZS. 100...500 optimisable parameters, that's what it was all about.
I do not understand the expression: 100...500. What does it mean? Please clearly define the task set for the participants. I don't think this has been done yet, thank you.
Please do not try to adjust to the championship conditions, whatever they may be - it will achieve nothing, because the problem will not be known to the algorithm. Algorithms should be versatile and be able to solve a wide range of optimization problems. Make your algorithm survivable in various conditions, focus on the number of optimized parameters from 100 to 500.
Take a look at the standard MT optimizer. It doesn't have any parameters that would allow you to customize it at all because it is universal and was designed to be so. If it had settings, we would receive a whole lot of complaints from users about its wrong settings and lack of help in setting up the optimizer. But you can't write a help for every particular task! That is why it is not configured. Every optimization task is unique and the user should be able to solve it without deep knowledge of the inner workings of the optimizer.
That is why there are still no clear limitations and "corridor" for the championship algorithms, because the algorithms will not know anything about the task! Make the algorithm universal, survivable.
If you are going to develop the algorithm from 0, you will need a lot of time, you won't be in time for the championship. It's better to take a ready-made algorithm, like in ALGLIB, and tweak it for you. At the same time you will gain deeper knowledge about how such algorithms work, and perhaps be inspired to write your own, unique creation in the future.
Good. Look, in order to explain the complexity of the task ahead of the participants in the championship, I will explain the order of the championship. And to emphasise a level playing field for everyone, including the organiser. I am currently doing absolutely nothing with my algorithm, I am not preparing for the championship, because it will not give me anything anyway, because I do not know the upcoming task.
1. Participants post their algorithms freely in a branch. From this point on, participants cannot change their algorithms.
2. Discussion and FF formation by participants begins. Participants propose their FF (perhaps trying to present such a FF, which his algorithm solves very well to increase their chances). For example, received 10 FFs. Then, these 10 FFs are submitted to one of MQ official representatives, who will randomly generate a sequence of these FFs, for example 1-2-3-5-8-2-3-9-10-1-2-5-5-7-6-....... (after that he will submit a free *.ex5 library with the FFs already compiled). Where number is number of FF. Each FF has two parameters, so it would be possible to look it visually in the form of 3-dimensional graph. Each FF has its own global max:
FF(f1(x1,x2); f2(x3,x4); f3(x5,x6); f5(x7,x8); f8(x9,x10); f2(x11,x12); f3(x13,x14); f9(x15,x16); (x17,x18); f10(x19,x20); f1(x21,x22); f2(x23,x24); f5(x25,x26); f5(x27,x28); f7(x29,x30); f6(x31,x32); ...)
x1,x2,x4,x4.... These are optimisable parameters, of which there can be between 100 and 500. Why is 500 parameters the ceiling? Because it's hard enough for FF, and fast enough to be counted - not all spectators have very fast computers that will be able to make sure the results of the championship are transparent.
The FF max will be the sum of maxima of these individual FFs and it can be calculated to be able to check and evaluate the algorithms.
So, I hope it is clear now, that in such conditions it is impossible to foresee and tweak your algorithm for a particular task in the hope of winning? The truly robust algorithm will win. I'm just looking forward to the start of the championship, I'm in complete ignorance of who will win, that is the intrigue! :)
I've never used tester optimization, so I'm unfamiliar with how it works. Putting someone else's algorithm to the championship is not for me. I won't be able to create a universal algorithm that solves any task in a year (or even in my lifetime). In the absence of understanding of the essence of the task set, I'm powerless. Conclusion - I will solve the problem that I understood from your explanations: FF is an analytical (mathematically speaking, the one that draws a curve on a graph) function. By passing values into it, in response I get values which are coordinates of points on a graph. Drawing a line through them yields a curve, with the proverbial peaks and troughs. Based on the logic of the obtained values, I look for highs and lows. The pictures you showed earlier also clearly show a surface with peaks. The discussion among the participants also contained an analogy with the surface and the peaks. Why have you now abandoned this analogy?
No, I didn't. That's the way it is. The FF at the championship will be a mix of FF competitors. If you take the FF individually, it can be represented as a 3-dimensional graph. But the FF of the championship cannot be plotted as a graph - it's multidimensional because. Everything is as I said before, nothing has changed.
The pictures above are simple examples for clarity, they are smooth. But we don't know what will be FF championship, some functions may be deliberately given discrete properties, not smooth, discontinuous, in the form of steps or holes, or flat horizontal surfaces. Therefore, if one imagines FF as simple three-dimensional plots as in the examples above, such representation would not be complete, to put it mildly.
I've never used tester optimisation, so I'm not familiar with how it works. It's not for me to put someone else's algorithm to the championship. I won't be able tocreate a universal algorithm that solves any problem in a year (or even in my lifetime).