Optimisation results differ from single tests on them - page 4

New comment
 
eugene-last:
Define the tf... One indicator, yes, is used. There tf: NULL, PERIOD_H1
Pretty much standard. And what or what else could be related to the tf?


Yes, maybe try it this way:

Supplement your code with the following - this is in the global variables

// Глобальные переменные
//
static datetime prevtime = 0;       // по ценам открытия

this is right after start

int start()    // -----------------------СТАРТ ЭКСПЕРТА--------------- 
{
  
 
   if(Time[0] == prevtime)   return(0);  //ждем нового бара
   prevtime = Time[0];                   //если появился новый бар , включаемся 
...

Test and opyt on TF M1 by opening price model - only for Expert Advisors with explicit control of bar opening...

Further, everywhere, in the indulators you use, in the EA itself, you should explicitly prescribe the working timeframes, as you consider necessary, for example,

double MA_1 = iMA(Symbol(),PERIOD_D1,Period_MA,0,MODE_EMA,PRICE_TYPICAL,1);
Later, write back here with the results of testing and optimisation...
 
Not all brokerage companies have enough history on M1, if anything, then try to test and optimize on a TF no larger than the one you explicitly stated in the SOV or INDICATORS, ie if you write "There tf: NULL, PERIOD_H1", then test and optimize on H1 by opening price model - only for EAs with explicit bar opening control...
 

Then contactthe technical support, wait for the shooting.

Here's the FAQ thread, it's usually on the front page of the forum. It's easy to find where to go for suspected tester or terminal errors.

You need to provide the following:

1. the source code of the Expert Advisor.

2. The full set used in testing/optimization.

3. and finally, clearly state the essence of the problem.

 
Mathemat:

eugene-last, then contact technical support, wait to shoot.

Here's the FAQ thread, it's usually on the front page of the forum. It's easy to find where to go for suspected tester or terminal errors.

You need to provide the following:

1. the source code of the Expert Advisor.

2. The complete set used in testing/optimization.

3. Finally, clearly state the nature of the problem.


It's getting late...

Looks like it's already gone...: "Even assuming that something is going wrong WITHIN the passes, well at least the very first pass in the optimisation should be identical to the single test?!
 Go shoot...."




 
Roman.:


It's too late...

Sounds like: "Even assuming that something goes wrong WITHIN the passes, well at least the very first pass in the optimisation should be identical to the single test?!

Go shoot...."

I removed the indicators, the result is the same - inconsistency.
I'm going to go through the whole process, I'll start with the first function and gradually add one function at a time until I come across the function that causes the inconsistency.
Waiting for your tech support, I'm used to what I know, but here .............
 
eugene-last:
I removed indicators, the result is the same - inconsistency.
I will go through all the steps, will start with the first function and will add them one by one, until I will come across the function, which causes the inconsistency.
Waiting for your tech support, I'm used to what I know, but here .............


One thing you should understand is that this is a NORMAL process of debugging cov code, especially if it has a twist, as they say.

Cover everything with printers - Print(); and in the tester in visualization mode by opening prices via F12 - step by step, bar by bar - track the contents of the "Log" tab of strategy tester, where printers all report the value of this or that parameter or variable... etc.

With the right approach, you'll get caught up and come out with your own error in the code!

Nevertheless, you should read all articles about strategy tester's work... :-)

 
eugene-last: I'm going to go through it all, starting with the first function and gradually adding one more function at a time until I come across the function that is causing the inconsistency.

That's the right approach.

I'm having a problem myself at the moment. Until I can be sure that I can't solve it, I won't write to tech support or the forum.

 
Roman.:
Not all brokerage companies have enough history on M1, if anything, then try to test and optimize on a TF no larger than the one you explicitly specified in the SOV or INDICATORS, ie if you write "There's a TF: NULL, PERIOD_H1", then test and optimize on H1 by opening price model - only for EA with explicit bar opening control ...

In this case, it is preferable to test on a lower timeframe than that prescribed in the indicators.

Otherwise, in this hour, the EA will only be able to close the position, and will only be able to open in the next hour, and that only if the conditions are still met.

m1 - m15 are the most suitable for testing an EA, working on n1, and it is even more important if the EA closes on tp and sl.

 
eugene-last:

Oh, and one last quirk. If you perform optimization several times, without genetics, just say 32 passes. So comparing reports of SEVERAL optimizations we see that the results coincide 100%.

You choose any pass, run it once and get a different result.

Even if we assume that something goes wrong between passes, well at least the first pass in optimization should be identical to a single test?!

Go shoot....

Whoever is destined to hang in a loop won't shoot himself.

here's another tip: it has been observed many times that similar cases with exactly the same test results are eliminated by restarting the terminal.

restart the terminal and get new/different test results.

 
mersi:

he who is destined to hang in a noose will not shoot himself.

here's another tip: it has been observed many times that similar cases with exactly repeated test results are eliminated by restarting the terminal.

restart the terminal and get new/different test results.

Delete the cache and get new results, I am aware of that
1234567
New comment
Reason: