Hello ladies and gentlemen. I decided, after all, to switch to MT5. Of course, there are a lot of new and interesting things. One thing was very frustrating, in the 64-bit version it is impossible to work without a genetic algorithm with iteration options of more than 100 million. Even if you ask a full search, with the number of options more than 100 million, the tester automatically turns on the genetic algorithm. In my case, I put the computer on for a few days and I am satisfied with a larger number of options. Is there a way to increase this number, or maybe someone knows how to contact the developer? I didn’t find any contacts on the MetaQuotes site.
Thank you all in advance.
Automatic switching to the genetic algorithm, how to deal with trouble.
Share
223231
What unique tasks each trader decides.
laryx
I think the point here will be that a much larger number of users will not understand why all of a sudden their optimization began to take much more time than it took before. It is easier to turn on the genetic optimizer on the machine than to explain to hundreds of users where the strange number of millions of passes comes from.
c00l777
That’s the point. It is emissions that work. Moreover, they have been working for 3 years. The result is stable, 600% per year.
laryx
I also do not understand what is the need for a complete search.
If there is a region of high values in the fitness function, then the genetic algorithm will find it and reach the maximum point.
If the maximum of the fitness function is not in the region, but simply in a single “outlier” – then, although the genetic algorithm can skip it (and a full search will find it) – it is unwise to use the “outlier” data.
c00l777
Thank you, the breakdown is clear. The problem is that often the tester finishes at night, and this is a waste of time. The computer was configured specifically. Core i9-7980 – 18 cores, 32 Giga memory, DDR4 frequency 2666 Mhz. As for the adviser, it is really simple MA/BB. It has been working since 2014, but the number of transactions is 10-15 per year, you will not earn much. On the previous computer, I managed to bring only 2 pairs to mind, testing takes 2-3 months. I hope to connect a dozen more on this.
renat
There’s even save settings and results of 100,000,000 passes (how many bytes per pass?) you may not have enough memory.
If the technique stops the frontal method, then you need to look for another smarter option.
If you have hundreds of millions/billions of passes at minimal cost, then rewrite your algorithm as:
for(int i=0;ifor(int j=0;jfor(int k=0;kand make a frontal pass.laryx
Divide your range into several, so that each of the sub-ranges includes less than 100 million variants. And iterate over the sequence.
I wonder what you have there for such a supercomputer, and a super-simple expert that 100 million options require only a few days of brute force ? Even if you bet for a week – you need to do 200 passes per second.