MetaTrader 5 trading strategy tester ideology: agents - page 9

 

:)) We need an auction - someone will want to sell cheaper - we need a market . :)

 

The terminal can use the resources of remote strategy tester agents by connecting to them over the network.

Here is what the first launch of a remote tester agent looks like (testing is managed by the trading terminal):

At the first launch, the agent manager offers to start several agents according to the number of logical processor cores, specifying the password for access and the desired ports. It is to these ports that the client terminal will connect when distributing tasks to remote testing agents.

After clicking the "Install" button, the services of agents will be installed and the manager shows them in the list:


From this window you can easily manage each copy of the remote agent.



The terminal has the option to use local agents (also by number of cores) - these are started without a GUI automatically with all parameters specified on the command line.

 

a question about neural networks...

Willneural network training also be paralleled or will it only be powered by its own power?

 

This will depend on the implementation of neural network training. The terminal itself does not provide neural network mechanisms in-house.

The MetaTrader 5 tester provides mass optimisation using genetic algorithms - just for them, multiple agents will speed up calculations by orders of magnitude.

Генетические алгоритмы - это просто!
Генетические алгоритмы - это просто!
  • 2010.05.25
  • Andrey Dik
  • www.mql5.com
В статье автор расскажет об эволюционных вычислениях с использованием генетического алгоритма собственной реализации. Будет показано на примерах функционирование алгоритма, даны практические рекомендации по его использованию.
 
When will we see the tester?
 
I don't know if I've heard this question before. Will there be a possibility of parallel testing of several instruments in the Tester? For synchronised testing of historical data? To test arbitrage, pair trading and multi-currency Expert Advisors?
 
maxis_tm:
I don't know if I've heard this question before. Will there be a possibility of parallel testing of several instruments in the Tester? For synchronised testing of historical data? For testing of arbitrage, pair trading and multicurrency Expert Advisors?
Yes, the tester is fully multi-currency.
 
Renat:

Protection mechanisms:

  • ...
  • only local agents can use the DLL with the appropriate permission in the terminal
  • passed expert code is never saved on the agent's disk, but is passed in a converted form impossible for dumping
  • agent does not know the Expert Advisor's name and does not save calculation results on disk (this minimizes the amount of information left after the calculation)
  • ...
Renat, what if optimizing Expert Advisor in the Strategy Tester applies the preliminary reading of prepared data from several text files #1.1, 1.2, etc. at every run, followed by writing some values obtained in file #2 at the end of that run? Will your agents be able to operate with such add-ons to the EA? P.s.: Personally, I like the idea of cloud testing very much - because it's a real solution to the problem of optimizing complex massive MTS on long sections of history, if genetics cannot be used due to the peculiarities of the robot. And the scepticism of some forum members is not relevant here at all
 

In the case of using writing to the same shared files, only one agent can be used:

  1. only one agent in sequential mode
  2. or implement its own system for assembling the results of the runs.
In principle, we can do a reverse transfer of a certain fixed file back to the control terminal. The EA can write any of its own results into it and then use it on the terminal side.
 
Renat:

... In principle, we could do a reverse transfer of a certain fixed file back to the control terminal...

yes... The task of transferring to/from agents the entire EA working environment including any set of files chosen by the EA writer, and all the more so that the results cloned at the agents would be assembled into a single output set using a key, and all this would not be FORMALIZABLE, because the environment in each particular case would be too individual. So your variant with predefined output file is the only practical solution.

That's why I was stuttering about printing to file on the agent side, with drag-and-drop results to the master: In addition to variables on the tab'Optimization results' of the tester, there is a need to have some more evaluation values - for example, some of those presented on the tab Report (take percentage of profitable deals), and if we're really desperate, we need indicators that are calculated on the Championships - it's worth something, if you know what I mean. Although this is the area of delights for the experts, of course. Yes, you're still counting yours...

And reading text files on the agent side is what I personally need to download values of tested parameters and emulate multicurrency on the basis of pre-prepared history. Yes, this last option is inheritance from monocurrency Tester of 4, but since this feature ate my whole carload of time, I stuck to it now, thankfully it has some other advantages, which will keep their relevance even in Tester 5.

Renat, will the manager terminal be able to send any number of text files to the Expert Advisor so it can read them? With writing - you have outlined the prospects, thank you. And as for reading, please specify what we can hope for

Reason: