A little surprised :) Thought I'd share and ask a NOT rhetorical question. - page 3
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
The optimiser is a search and the tester is a check.
The optimiser says - it will be positive at such and such parameters, and the tester confirms - yes, that's right. :)
So "realism" can be different for the tester and for the optimizer?
Absolutely right. But the overall picture you will see is correct. Roughly speaking, when you run in the tester optimisation by opening prices, and then individual results by ticks, the picture doesn't change much, does it? Exactly the same with the optimiser. Only it is not always optimal to simplify the optimization by the "open prices" model only.
And as was rightly mentioned, the fastest optimizer is the for loop + use of EA's specific features. So, it is not a universal approach at all.
Obviously, we do not need to check the margin and other parameters all the time in the optimizer, as it is done in the tester. The optimiser can give a super-fast estimate of the idea, which can then be honed in the tester. Or the optimizer, as it is now, gives an almost absolute simulation of the real thing at each step, which is very slow and understandably not 100% in demand.
For developers to make a fast optimizer, there has to be a mode where a bunch of checks don't have to be involved. This is a kind of exploratory mode of an idea.
I.e. the optimizer can be intended for research and it can be used to super-hone an already prepared idea. The second one is implemented.
I don't understand why developers should write the optimizer as a super-fast variant. Someone will be unsatisfied with it anyway. Those who need the super-fast version can easily write the for loop for themselves taking into account their trading ideas.
To optimize my Expert Advisor, I use the self-written optimizer-tester. It is more than 100 times faster than the MT4 tester. The results differ less than 1%. After the optimization on my tester, I am perfecting trading features on the MT4 tester. Nothing prevents anyone from implementing such a scheme.
In short, if you want a fast optimizer, write it yourself.
That's what draws everyone to "search" by the optimizer . Freeloaders ))
"We're not freeloaders, we're partners",
Putting the slogan "manual labour on the shoulders of machines" into practice :о)
Absolutely right. But the overall picture you will see is correct. Roughly speaking, when you run in the tester optimisation by opening prices, and then individual results by ticks, the picture doesn't change much, does it? Exactly the same with the optimiser. Only it is not always optimal to simplify the optimization by the "open prices" model only.
And as was rightly mentioned, the fastest optimizer is the for loop + use of EA's specific features. So, it is not a universal approach at all.
Obviously, we do not need to check the margin and other parameters all the time in the optimizer, as it is done in the tester. The optimiser can give a super-fast estimate of the idea, which can then be honed in the tester. Or the optimiser, as it is now, gives an almost absolute simulation of the real thing at each step, which is very slow and understandably not 100% in demand.
In order for developers to make a fast optimizer, there has to be a mode where a bunch of checks don't have to be involved. This is a kind of exploratory mode of an idea.
I.e. the optimizer can be intended for research and it can be used to super-hone an already prepared idea. The second one is implemented.
I don't understand why developers should write the optimizer as a super-fast variant. Someone will be unsatisfied with it anyway. Those who need the super-fast version can easily write the for loop for themselves taking into account their trading ideas.
To optimize my Expert Advisor, I use the self-written optimizer-tester. It is more than 100 times faster than the MT4 tester. The results differ less than 1%. After the optimization on my tester, I am perfecting trading features on the MT4 tester. Nothing prevents anyone from implementing such a scheme.
In short, if you want a fast optimizer, write it yourself.
"Are you Brute? (с)
I'm surprised, with your scrupulousness and meticulousness in questions of exactly what methodology to search for an "optimizer" ... Let it be a question about a chicken and egg and everyone with a clear conscience will have his opinion ))
That's the point, it's not good to reduce everything to Mathcad + Matlab. Sometimes it is useful to investigate a simple idea already in the form of buy-sell, without even understanding the essence to the end. And only an optimizer can investigate it quickly and easily. That is, a thorough research does not always come before the optimizer. Sometimes, without research, the optimizer comes right away as part of the research. It's almost butter-and-oil, but I think it's understandable.
The optimizer can be written in C++ and MQL5 and Matlab+CUDA. All you need is for it to satisfy your needs.
I have described my scheme: I have a working idea, and I'm sure of it. And there's even no need to investigate it. But I need optimization to find the optimal parameters and to diversify by them. We take several dozens of FI, not small stories on them, and run it on each of them. As I said, speed is > 100 times faster than MT4 tester (by opening prices), and error is less than 1%. Of course, I've optimised everything there, no near-checking or complicated history work. I know my idea, that's why I wrote such a fast version. MT4 would count for a week. In my variant it is an hour. The time difference is huge, while the error is miserable. I further improve it using the MT4 tester.
Everyone needs a fast optimizer for their needs. The developers have made it so that my optimizer with the 1% error just "beat" the speed having run it on a hundred or two testers-agents and got almost zero error. And what's more, they give full versatility.
The question is simple: universality + scalable hardware power vs. non-universal algorithmic optimization. Obviously, the developers have chosen the only right way.
I'm more interested in the official and public result of comparing the speed of an MT4 tester with that of an MT5 tester.
When you easily argue that your own quick tester is like a MetaTrader tester, don't forget to point out that you have virtually no analytics (indicators) in your own quick testers at all. You basically can't test and achieve 99% accuracy of an in-house tester.
In fact, adding any indicator to the custom tester will immediately set the speed at the level of the regular tester or even slower. For example, EURUSD has about 51.5 million ticks over 11 years. Insert the calculation of any indicator (of course, economical one) inside this cycle and the speed will decrease by several orders of magnitude. But because custom testers rarely use indicators, it turns out that they don't face the speed drop so clearly.
Speaking of speed, keep in mind that the in-house tester is very well optimized and can very effectively spin multi-million cycles, simulating bars. We have optimised the tester's processes well.
Another interesting point is that the custom tester has no possibility to visualize the results both in the form of a chart and in the form of showing of deals on the chart.
When you easily argue that your own fast tester is like a MetaTrader tester, don't forget to point out that you have virtually no analytics (indicators) in your own fast testers at all. You basically can't test and achieve 99% accuracy of an in-house tester.
In fact, adding any indicator to your own tester will immediately set the speed at the level of the regular tester, or even slower. The lack of possibilities to visualize the results both in a chart and in the form of trades on a chart adds a special piquancy to the custom tester.
The indicator concept itself seems (to me, of course) wrong in the case of a fully automated system. I am not saying this as a developer of platforms with 10 years experience, but as a practitioner of fully automated trading, who has been successfully working for only 5 years. Roughly speaking, at least all indicators are transferred to the Expert Advisor. No offence, MQL5 Wizard has only one useful thing. Now everyone can be sure that the indicator concept is rubbish, no matter how complex the combinations of indicators may be.
Well, the major technically advanced market participants do not trade using indicators. Because the indicators are kindergarten of the 20th century. Just for humanitarians and lazy technicians.
Speaking about speed, we should keep in mind that the built-in tester is very well optimized and can effectively spin multimillion cycles simulating bars. We have well optimised the tester's processes.
The indicator concept itself seems (to me, of course) wrong in the case of a fully automated system. I am telling you this not as a platform developer with 10 years of experience, but as one who has successfully practiced fully automated trading for only 5 years. Roughly speaking, at least all indicators are transferred to the Expert Advisor. No offence, MQL5 Wizard has only one useful thing. Now everyone can be sure that the indicator concept is rubbish, no matter how complex the combinations of indicators may be.
Well, the major technically advanced market participants do not trade using indicators. Because the indicators are kindergarten of the 20th century. Just right for humanitarians and lazy technicians.
Maybe so, but without a comparative analysis it is just words.Oh, that's it, we play without indicators only on pure bars?
Then the "100 times faster" is out of the question. Most likely one can say "the regular mt5 tester is faster on single tests and several times faster on optimization".
In fact, a tester without indicators is not serious at all. You may not use indicators, but a tester without the effective support of indicators cannot be taken seriously.
So that's it, we only play without indicators on clear bars?
That's right.
Then "100 times faster" is out of the question. Most likely one will be able to claim "in-house mt5 tester is faster on single tests and several times faster on optimization".
I think anyone would be happy to have an official and open comparison of speeds of MT4 and MT5 testers on non-syndicator EAs. That's where we will see the real speed of the tester and not the brilliant architectural and algorithmic optimization of crutches-humanitarians in the form of indicators.
In fact, a tester without indicators is not serious at all. You may not use indicators, but a tester without the effective support of indicators cannot be taken seriously.
We have different approaches. You - the platform developer. I'm an "advanced" MTS-nik. Who are your clients mainly? - DC and humanitarian traders (+ lazy technicians). They are the ones you rely on in your developments, which is correct. In my case my client is the market. And the indicator concept applied to the market is nonsense. You are fed by your clients. I am fed by mine. Both are right.
From my point of view, it is not good for an "advanced" MTS-nickel to reproach developers that they do not create a perfect toolkit for him. They should also ask for money. Everyone does what he needs to do.
Question to the developers:
Will the Expert Advisor with indicators and the same Expert Advisor, but with transferred indicators in its code ("all in one"), be different in terms of speed of execution in the tester? In which direction?
Most likely, there will be no definite answer to this question. But nevertheless I ask you to somehow more or less elucidate this question.