Your symbols and your datafeeds in Metatrader 5 - page 7
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
In genetics, of course, these areas are also visible, if I were to overshoot, I would still expand the parameters for analysis there, in genetics I would miss this point with an 80% probability
And there's a border on the left that needs to be investigated, which genetics doesn't show
In genetics these areas are of course also noticeable, but I would still expand the parameters for analysis there, in genetics I would miss this point with a probability of 80%.
The genetic algorithm in trading system testing is not an endpoint, but a method for finding directions for more detailed research.
The algorithm is usually as follows:
If someone thinks that a genetic or any other fairy tale optimizer should immediately give clear results it means that he/she does not understand the process at all.
In increments of 0.01
The genetic algorithm in trading system testing is not an end point, but a method of groping for directions for more detailed research.
The algorithm of operation is usually as follows:
If someone thinks that a genetic or any other fairy tale optimizer should give us clean results immediately, it means that they don't understand the process at all.
Ok, agreed, convinced, will use as you recommend :-) I, with your help, understood, genetics allows to choose the most promising areas in a gigantic range. and then explore them
Two questions remain.
1) Why is the genetic algorithm switched on PRECISELY by the condition of N number of variants ?
2) What does the terminal download when working in the mode Mathematical Calculations? While I was generating nice pictures here, the terminal has already downloaded 40 meters - it doesn't download so much even when I use Expert Advisors in the Strategy Tester.
Tick, bar and tumbler control is also available.
If, of course, the particular datafeed supports it.
I'm going to roll genetics :-)
OK, I agree, convinced, I will use as you recommend :-) I, with your help, realised, genetics allows you to select the most promising areas in the gigantic range. and then explore them
Two questions remain.
1) Why is the genetic algorithm switched on by the condition of N number of variants ?
Because it physically doesn't make sense to do a full enumeration beyond the boundary. For 32 bit platforms it is 1,000,000 passes, while for 64 bit platforms it is 100,000,000 passes.
Is it so hard to understand? Well, you will never wait for 100 000 000 seconds. You will not, and never will.
2) What will the terminal download when working in Mathematical Calculations mode? While I was generating nice pictures, the terminal already downloaded 40 meters, it is not downloading so much even when I use Expert Advisors in the Strategy Tester.
Function Z = cos(1.5*x)*cos(1.5*x) + sin(2.25*y) + cos(3*x*y); where X and Y are -3 to +3
I'm also wondering how to find its maximums in MT5.
As for the method - the idea is from an article on hubra, the implementation is in matlab and in C#.
The problem with the proofs is that it's hard to give them, as the author has no practical experience. Unlike MetaTrader developers who have been doing it for many years.
Into the granite! It is your catchphrase, there is no sense to move against it, just about this senselessness I stated at once. You don't need proof, you don't even question your own rightness. I'm sure the appearance of tick data and custom feeds, from your point of view, does not resonate with the granite phrase, when you were asked about it and proved the feasibility for N years. Experienced as you are!
Unfortunately, you just cited known basic theoretical points.
And I asked a specific question "What don't you like about GA? It doesn't find areas of solutions for you?". Of course it does, and no worse than any other method. And it gets out of local holes better than annealing. But the most important thing is that it solves your problems efficiently.
You haven't read the article at all, which is why you mention this "churn". But the article doesn't use it in the slightest. "Haven't read it, but I condemn it" - you can't read experience, can you? Specifically, I am not satisfied that GA does not find convergent (not random) local extrema. Therefore its suitability for TC optimization is doubtful, to say the least. So many heuristic algorithms were invented for a reason. And there is no best heuristic algorithm. Each task has its own best one. So GA for this very TC optimization task is, unfortunately, far from being the best. The arguments are given in the article for those who want to understand what we are talking about.
Because it physically doesn't make sense to do a complete overrun over the edge. For 32 bit platforms it's 100,000,000 passes and for 64 bit platforms it's 1,000,000,000 passes.
Is it so hard to understand? Well, you will never wait for 1 000 000 000 seconds. You never will.
This is traffic between local agents and terminal. It shows up as network traffic, being in fact network traffic (even if within localhost).Okay, I give up :-)
One last try, purely out of curiosity, how much would 1,000,000,000 passes cost if you leverage the entire remote agent service's existing network? I haven't used remote agents in a while, don't know the prices.
And how much time would it take the whole network of remote agents to make 1,000,000,000 passes, at least approximately?
Yep, the results were posted without me. You can clearly see that the algorithm described in the article finds local extrema almost immediately and of a much higher quality than the standard GA.
No, it does not.
It is not that simple function Z = cos(1.5*x)*cos(1.5*x) + sin(2.25*y) + cos(3*x*y); where X and Y are from -3 to +3 that I discussed in the article as suggested above.
Not only that, but the author of that article actually built a huge bicycle (which is good for self-study, of course), but apparently he sharpened the search optimization for his task. This optimization will most likely cause problems (increase in amount of calculations) with other tasks.
There are also important metrics left out - how many passes were actually made in heuristic mode compared to a full overshoot. For example, in the MT5 example above we got 8,700 in genetics and 361,201 in brute force. There is a suspicion that the author's own heuristic optimized variant actually spent much more passes to complete the results.
The number of passes is very important, since rarely any strategy meets the deadline of seconds. The difference between our GA with 10,000 passes and another with 30,000 passes results in waiting for an extra 20,000 passes * time of pass, which is very long. Our GA is specifically optimised for the fastest possible miscalculation. Usually ours is sufficient for 10,000 to 12,000 passes regardless of the total size of the search field. This means that any depth of search can be carried out roughly in 10,000 passes. From here on, we have a head in the hands and explore more accurately.
By the way, in MetaTrader 5 the author didn't have to spend months on writing his own engine, and you can immediately get the results by pressing the button. And in 2D/3D you could spin it in different projections.
Into the granite! That's your catchphrase, there's no point in going against it, that's the pointlessness I made right away. You don't need proof, you don't even doubt you are right. I'm sure the appearance of tick data and custom feeds, from your point of view, does not resonate with the granite phrase, when you were asked about it and proved the feasibility for N years. Experienced as you are!
My work is visible to all. Yours is not visible, unfortunately.
If you think that MT4/MT5 genetic engines have passed me by, you're very much mistaken.
You haven't read the article at all, which is why you mention this "churn". But the article doesn't use it in the slightest. "Haven't read it, but I condemn it" - you can't read experience, can you? Specifically, I am not satisfied that GA does not find convergent (not random) local extrema. Therefore its suitability for TC optimization is doubtful, to say the least. So many heuristic algorithms were invented for a reason. And there is no best heuristic algorithm. Each task has its own best one. So GA for this very TC optimization task is, unfortunately, far from being the best. The arguments are given in the article for those who want to understand what I mean.
Do I have to write in each case a legally accurate phrase like "monte carlo methods, annealing, etc."?
I wrote this phrase once above. Then I summarized "heuristic methods", pointing out that even the regular genetic algorithm can be turned into another method by playing with custom criteria and stimulating the engine to keep calculating.
The arguments in the article are only of a general nature. But the stool created for one strategy (take a look at the screenshots) and exactly the same method of tweaking shows that there must be scaling problems when moving to a universal calculation platform, which MT5 + MQL5 is.
Yes, the article is good as a demonstration of what the programmer has done. But in practical terms, it is not the best ten thousandth bicycle invented by yet another beginner.