Machine learning in trading: theory, models, practice and algo-trading - page 1177

 
Grail:

It looks hard, it was already said somewhere above that one of the signs of low overfits is exactly the similarity of the graphs of equity on the Lerning and Test, actually the same logic on the classification/regression, and the equity as a consequence.

It was said here a lot, but as a rule without proof, so not to repeat, publish your own experience, at the same time and will prove to all that the market is unchanging
 
Grail:

The devil is in the details :) For example, for some reason it turns out that on different testers, the results of the same strategies, on the same data come out different, sometimes significantly, but the right (closest to the truth), someone is one.

To avoid this dilemma, you should use your tester with a fully controlled structure. In this case there is full confidence that the tests will not differ significantly from the real work.

 
Ivan Negreshniy:

The engine is integrated into a large project, there are many megabytes of source code in several languages, some interpreters, in addition to python and p, there are also scripting java and pascal.

And if you are interested in the principle and example of Python code execution, which I use, I offered it here a long time ago.
https://www.mql5.com/ru/forum/86386/page553#comment_6302133

Thank you, the principle is clear.

I went the other way - the terminal separately, Python separately. The exchange of CSV strings through TCP client-server. This has been implemented for Lua, not for MQL yet, but no problems are expected.

 
Grail:

It looks hard, it was already said somewhere above that one of the signs of low overfits is exactly similarity of equity graphs of Lerne and test, actually the same logic in classification/regression, and equity as a consequence.

Well, I also wrote that in that case it isn't important which side is the tray and which side is the test

as stated here by people who have never used the MO anywhere, and in general, the path is not very far (just kidding, in fact I'm kind)

 
Yuriy Asaulenko:

Thank you, the principle is clear.

I went the other way - the terminal separately, Python separately. The exchange of CSV strings through TCP client-server. I have implemented it for Lua, not for MQL yet, but no problems are expected.

The main thing is reliability and speed, what would be possible to use in the Strategy Tester, for example.

As for me I may run mine even with realtime network training, it works tolerably well.
https://www.youtube.com/watch?v=73iic_vMiU8&feature=youtu.be

 
Ivan Negreshniy:

The main thing is reliability and speed, so that for example in the same tester could be raced.

Here I run mine, even with realtime learning network, works tolerably.
https://www.youtube.com/watch?v=73iic_vMiU8&feature=youtu.be

o normas, I have the same, only the graph grows in the other direction :)

 
Maxim Dmitrievsky:

Oh normas, I have the same, but the graph grows in the other direction :)

This is an example of interaction between EA and python console in the tester, when at each step there is not only a network response calculation, but also periodic additional training.

P.S. The graph is not indicative of another pass, there the console lives separately and can simultaneously receive commands from several sources on the shared network

 
Ivan Negreshniy:

The main thing is reliability and speed, so that, for example, in the same tester could be raced.

Here I am running mine, even with realtime network learning, works tolerably well.
https://www.youtube.com/watch?v=73iic_vMiU8&feature=youtu.be

What kind of python lib is used?

 
Yuriy Asaulenko:

What kind of python library is used?

If about the language lib, then any of the installed together with application libraries, just specify the link in the settings of the engine, the corresponding version x32 or x64.
 
Ivan Negreshniy:
If the language lib we are talking about, then any of the installed together with application libraries, only link to specify in the settings of the engine, the corresponding version x32 or x64.

Nah, I mean, what do you teach? - The packet-liba? I thought it was about the network.