Discussion of article "Controlled optimization: Simulated annealing"

New comment
 

New article Controlled optimization: Simulated annealing has been published:

The Strategy Tester in the MetaTrader 5 trading platform provides only two optimization options: complete search of parameters and genetic algorithm. This article proposes a new method for optimizing trading strategies — Simulated annealing. The method's algorithm, its implementation and integration into any Expert Advisor are considered. The developed algorithm is tested on the Moving Average EA.

Implementation of the algorithm will require two new classes, which should be included in the optimized Expert Advisor:

  • AnnealingMethod.mqh class — contains a set of methods implementing separate steps of the algorithm;
  • FrameAnnealingMethod.mqh class — contains methods for operation of the graphical interface displayed in the terminal chart.

Also, operation of the algorithm requires additional code to be included in the OnInit function and adding the functions OnTester, OnTesterInit, OnTesterDeInit, OnTesterPass to the EA code. The process of integrating the algorithm into an expert is shown in Fig. 2.


Fig. 2. Including the algorithm in the Expert Advisor

Author: Aleksey Zinovik

 
MetaQuotes Software Corp.:

New article Controlled optimization: Simulated annealing has been published:

Author: Aleksey Zinovik

I haven't started to work with mt5 yet but I guess that (may be with some changes and/or conditional compilation) I will be able to use this in the strategy tester of mt4 as well.

The genetic optimization of the mt4 has one little problem that may as well exists for mt5 too and may be as well for your approach.

If I check in OnInit() the parameter set and found that an actual set is not worth to be checked I return from OnInit() with INIT_PARAMETERS_INCORRECT. The optimization is not performed which saves time, ok - but:

  1. the genetic algorithm of the mt4 regards this as a valid pass through
  2. it therefore increases the counter so that
  3. the genetic algorithm stops a lot earlier than it should (I think its criterium to stop is a mixture of number of runs and the change of the target value).

Have you took care this situation: Returning from OnInit() despite there is no error (like: file not found,..) just because the parameter setup should not be tested which in this case should not increase the the number and decrease the temperature?

Anyway thank you for this interesting article!

Gooly

 
Carl Schreiber:

I haven't started to work with mt5 yet but I guess that (may be with some changes and/or conditional compilation) I will be able to use this in the strategy tester of mt4 as well.

The genetic optimization of the mt4 has one little problem that may as well exists for mt5 too and may be as well for your approach.

If I check in OnInit() the parameter set and found that an actual set is not worth to be checked I return from OnInit() with INIT_PARAMETERS_INCORRECT. The optimization is not performed which saves time, ok - but:

  1. the genetic algorithm of the mt4 regards this as a valid pass through
  2. it therefore increases the counter so that
  3. the genetic algorithm stops a lot earlier than it should (I think its criterium to stop is a mixture of number of runs and the change of the target value).

Have you took care this situation: Returning from OnInit() despite there is no error (like: file not found,..) just because the parameter setup should not be tested which in this case should not increase the the number and decrease the temperature?

Anyway thank you for this interesting article!

Gooly

I don't check the correctness of the parameters and don't interrupt the OnInit() function if the parameters are not correct. In the OnTesterInit() function, parameters whose values need to be optimized using the Strategy tester are disabled from optimization. At each new iteration, the parameters are read from the file, in the OnTester() function, new parameter values are written to the file. This makes it possible not to use the parameter values generated by the Strategy tester, to independently output the necessary parameters to the OnInit() function.
Sorry for my English
 
Aleksey Zinovik:
I don't check the correctness of the parameters and don't interrupt the OnInit() function if the parameters are not correct. In the OnTesterInit() function, parameters whose values need to be optimized using the Strategy tester are disabled from optimization. At each new iteration, the parameters are read from the file, in the OnTester() function, new parameter values are written to the file. This makes it possible not to use the parameter values generated by the Strategy tester, to independently output the necessary parameters to the OnInit() function.
Sorry for my English

I understand now - thank you! I just have to check the plausibility of the parameter setup of the next iteration right before this setup is written to the file for the next run. So this way 'my problem' can be avoided!

Thanks for this idea and the article!

 

Hi ,

You said '' drawbacks can be eliminated by developing a universal module, which would include various algorithms for optimizing the expert parameters''.

Can you give more details about the universal module ?What are the other  algorithms for optimizing the expert parameters?

 
nevar:

Hi ,

You said '' drawbacks can be eliminated by developing a universal module, which would include various algorithms for optimizing the expert parameters''.

Can you give more details about the universal module ?What are the other  algorithms for optimizing the expert parameters?

The article shows how to connect a new optimization algorithm to the strategy tester. Similarly, adding new methods to a class AnnealingMethod.mqh or creating a new class, you can connect other algorithms, for example, ant algorithms (Ant colony optimization). I plan to test the work of such algorithms and share result.

New comment
Reason: