MetaTrader 5 Strategy Tester and MQL5 Cloud Network - page 26

[Yedelkin]

notused:

With genetics, it's understandable - waiting for the rest of the population to return the result.

Ah, so that could be the point. Then with fast local agents working with "clouds" is contraindicated?

[Valerii Mazurenko]

Yedelkin:
Ah, so that could be the point. Is cloud computing contraindicated for fast local agents, then?

It depends on how many local/remote agents you have - if you have a dozen or two, the cloud is unlikely to make a difference, but if you have just a couple, the cloud will be faster. For experts who test relatively "fast", the cloud will help even when there are a lot of local/remote agents

[Alexey Da]

notused:

Still, something has to be done about it:

Pay attention to the response returned by the cloud. It seems that not a single correct response has been received since yesterday (and the passes are 100% with transactions and no errors). The usefulness of the network in such a case is 0. And the case - 12 instruments since the beginning of the year. Although a couple of days ago at times there were correct results, but today - none at all.

In light of the fact that the cloud will soon be paid, I would not want to throw money to the wind without any result (in the test mode I got $7 for yesterday and today for using the network, although, again, I did not get anything from it, and used the network only yesterday evening, because it was not working during the day).

+ sometimes the cloud will go into a state of being finished when it's full(!). How can this be when there are still a few thousand passes left?

Have you tried running a single pass on such results on a local agent?

Can you describe startup parameters, list of agents, optimization parameters in your request to Service Desk + add an expert (at least ex5)? The logs would also be very helpful.

[Andrey Sharov]

alexvd:

Have you tried to run a single pass with these results on a local agent?

Can you describe launch parameters, list of agents, optimization parameters in the application to Service Desk + attach expert (at least ex5)? And some logs would be really helpful.

The question is not for me, but yes, I've tried it. On local agent the test passes without errors. There is an assumption that the problem is related to timer (EST = EventSetTimer(3600)), because in my case critical runtime error occurred at the beginning of testing (testing time - 0 seconds). Literal text of OnInit (I can't insert it as source for some reason, IE8):

int OnInit()
{
Print("hh=",hh," mm=",mm," MLEVEL=",MLEVEL," TP=",TP," SL=",SL," TS=",TS);
Print("OnInit begins SymbolsTotal: ",SymbolsTotal(true),"/",SymbolsTotal(false));

// tumbling for initialization in the tester
for(uint i=0; i<maxsym; i++)
{
SymbolSelect(SymbolName(i,false),true);
}

EST = EventSetTimer(3600);
if (!EST) {
Comment("Error: EventSetTimer= ", GetLastError();
Print("Error: EventSetTimer= ", GetLastError();

}

for(uint i=0; i<maxsym; i++)
{
ea[i].Init(SymbolName(i,true),PERIOD_M1); // initialize expert
Print(i,SymbolName(i,true));
}
// initialization example
// ea.Init(Symbol(), PERIOD_M5); // for a fixed timeframe
// ea.Init("USDJPY", PERIOD_H2); // for a fixed symbol and timeframe

Print("OnInit ends SymbolsTotal: ",SymbolsTotal(true),"/",SymbolsTotal(false));

return(0);
}

[Serj]

Can anyone tell me how to do this?

I'm trying to speed up optimisation, I don't want (for example) results with a drawdown greater than 30%.

What I'm doing is this:

input double Max_DD_EQ = 30;

double max_eq,max_dd,dd_percent;
bool dd_valid=true;

//+------------------------------------------------------------------+
//| Timer function                                                   |
//+------------------------------------------------------------------+
void OnTimer()
  {
   if(!dd_valid) return; 
   if(MQL5InfoInteger(MQL5_OPTIMIZATION) || MQL5InfoInteger(MQL5_TESTING))
     {
      max_eq=MathMax(max_eq,AccountInfoDouble(ACCOUNT_EQUITY));
      max_dd=MathMax(max_eq-AccountInfoDouble(ACCOUNT_EQUITY),max_dd);
      dd_percent=max_dd/(max_eq*0.01);
      if(dd_percent>Max_DD_EQ) { ClosAllPos(); dd_valid=false; return; }
     }
// дальше идет блок открытия (закрытия) поз

I want the run to end after exceeding the allowable drawdown.

But the tester, after reaching the maximum drawdown, counts as before. I do not want empty (useless) runs.

Any other options? Or to the developers?

Or shorter: How do I forcefully terminate the run? (In MT4 there was such a thing.)

[Igor Volodin]

ExpertRemove();

?

[Serj]

Vigor:
?

Thank you so much! It worked!

Checked it out. It went a lot faster now.

[Valerii Mazurenko]

alexvd:

Have you tried to run a single pass with these results on local agent?

Notused:

I run a single pass on local kernel - everything is fine (I've tried many times, with different parameters). I suppose that perhaps the network agent either ran out of disk space or memory is not enough (multicast) or something else.

alexvd:

Can you describe the startup parameters, list of agents, optimization parameters in your request to Service Desk + attach an Expert Advisor (at least ex5)? And the logs would be very helpful.

Sent (#232346). Sorry no logs - they get overwritten by the terminal. But I think you'll have no problem reproducing it.

Ashes:

The question is not for me, but yes, I've tried it. On local agent the test passes without errors. There is an assumption that the problem is related to the timer (EST = EventSetTimer(3600)), because in my case critical runtime error occurred at the beginning of testing (testing time - 0 seconds).

Indeed, the timer applies. Only in my case you can see that the network agents work for about 5 minutes on the calculations, and then give an empty result.

[Slava]

notused:

how to fight it?

Sometimes I get responses like this from network in packs (Critical Runtime Error). Running pass on local kernel - everything is fine (I tried it many times, with different parameters). I suppose, that probably the network agent either ran out of disk space or memory is not enough (multicast) or what else. But, it's a shame that a generally profitable passage is excluded from further consideration. Although, on the other hand, I feel that the problem is not solvable in principle (how do I know that the problem is with the agent and not with the ekspert?) . Or maybe a setting "Do not use agents which returned Critical Runtime Error in last 24 hours on your account"?

Not critical for me (enough remote cores), but generally an inconvenience.

We are in the process of treating this problem.

[Yedelkin]

Here's a question. I have an 8-core Core i7, and after I've finished preparing for the championship, I don't really need it. Unless I want to play a game or something for half a year. In order not to stand idle, it would be possible to transfer some power to a cloud, but I wonder whether the computer will cool down in the same way as it would cool down if I optimise it locally. Of course, a computer could survive 5 days of continuous optimization with constant overheating, but still I had to keep an eye on it, cool it down, etc.

..The monetary component does not interest me. If the income from the cloud will be a penny, you can do it for free - NiVapros.