Discussing the article: "Using optimization algorithms to configure EA parameters on the fly" - page 2

[Andrey Dik]

Aleksey Vyazmikin #:

Interesting article, it should raise interest in the series of articles you publish!

Of course, the disadvantage of the proposed implementation is the lack of universality of the approach, i.e. it is necessary to completely rewrite the existing Expert Advisor and introduce a virtual tester with many functions into it. Of course, the plus of the approach is the acceptable speed of work due to virtual indicators.

Thank you for your comment.

There are minuses of this implementation, of course. But in this article I wanted to focus more on the concept of using optimisation algorithms proper. In fact, the schemes of using algorithms can be very diverse, the universal architecture of algorithms allows implementing any idea where something needs to be searched or optimised.

As for self-optimisation as a process of running the EA strategy on history, it would be ideal to have a standard MQL5 function like HistoricalSelfRun (startDate, endDate, params), which runs the EA on history in a lightweight version of the tester. Now you can run the tester from the command line, but this is a separate process and such solutions are not suitable for the market.

Have you tried to use not chart ranges for optimisation, but sets of such previously selected settings for each indicator/predictor? This approach significantly reduces the search area, but I understand that not all algorithms will work properly, as there is no smooth change from parameter to parameter, or is it not crucial?

I understand what you are saying, I think you mean that changing opt-parameters in some algorithms should be done by the algorithm itself and outside interference is undesirable? Some algorithms really don't like it if you interfere in their work from the outside by giving them coordinates, but most algorithms are tolerant of such interference. I am preparing an article covering the behaviour of algorithms in such a scenario. But yes, sometimes it can be very useful to start an algorithm from a certain position in the search space.

Are there any planned articles on other applications of algorithms, along the lines outlined in this article:

• Portfolio Management. Optimisation algorithms can help determine the optimal asset allocation in a portfolio to achieve given objectives. For example, optimisation techniques such as the Mean-Variance Optimisation (Mean-Variance Matrix) can be used to find the most efficient set of assets given expected returns and risk. This may include determining the optimal mix of stocks, bonds and other assets, as well as optimising position sizes and portfolio diversification.

• Selecting the best trading instruments. Optimisation algorithms can help in selecting the best trading instruments or assets to trade. For example, you can use optimisation algorithms to rank assets based on various criteria such as return, volatility or liquidity.

The field is unploughed on optimisation and these topics are very interesting. I hope, yes, to pursue these topics.

[Andrey Dik]

LUIS ALBERTO BIANUCCI neural networks or self-learning) should go, with a logical strategy, and we add this, and it's really beneficial, self-optimisation, this is what will bring us closer to human intelligence mode, because it allows us to make decisions in the short term, and based on what's happening in the moment. the moment, having the best, instantaneous conditions. moment having the best, instantaneous conditions.

+++

[Roman Poshtar]

Thank you. I'll double-check all the algorithms. Which one do you think is best?

[Andrey Dik]

Roman Poshtar #:
Thank you. I'll double-check all the algorithms. Which one do you think is the best?

What exactly do you want to double-check?

Which algorithm is better is for everyone to decide for themselves, all algorithms have both advantages and disadvantages, I have tried to show the pros and cons. Besides, there is a rating table, which speaks volumes.

[Rorschach]

You asked about distributions once, interesting to check the stratification

https://www.mql5.com/ru/forum/86386/page3395#comment_51967665

https://habr.com/ru/articles/496750/

[Andrey Dik]

Rorschach #:

You asked about distributions once, interesting to check stratification

https://www.mql5.com/ru/forum/86386/page3395#comment_51967665

https://habr.com/ru/articles/496750/

Please explain the context.

[Rorschach]

There was a question there, is there any interest in testing the effect of different distributions on optimisation algorithms, or something like that.

[Andrey Dik]

Rorschach #:

You asked about distributions once, interesting to check the stratification

I don't understand, where and when did I ask?

Any optimisation algorithm is based on playing with probability distributions. Therefore, distributions play a key role in the functioning of algorithms.

If you are talking about the uniformity of HF generation, then in relation to optimisation it is probably more appropriate to talk about the effect of the quality of the HF on the search properties. And this topic is in the plans for coverage.

Distributions were touched upon in this article.

[Rorschach]

Andrey Dik #: on the influence of the quality of the DST on the search properties.

That's a possibility.

The question was in Saber's cart, now it is not there.

[Andrey Dik]

Rorschach #:

It could be this.

The question was on Saber's cart, it's not there now.

Well, if there is an interest in the topic of the influence of the quality of HCS on search engine optimisation algorithms, then an article on this topic will be useful. I myself am interested in this question, the answer to which is not obvious.