Discussing the article: "Creating a mean-reversion strategy based on machine learning" - page 3

New comment
 
Maxim Dmitrievsky #:
Models take a long time to count.

How to add new attributes to onnx? For example, date - month, day, hour and minutes, individually 4 attributes.

[Deleted]  
Evgeni Gavrilovi #:

How to add new attributes to onnx? For example, date - month, day, hour and minutes, individually 4 attributes.

In onnx or in the function that generates the attributes? Pipeline itself is not converted to onnx in the article.

The get features function generates features, you just need to make changes to it.
 
I could not understand one point, theSavitsky-Golei filter itselfis not in MT5, then how new data are obtained?
 
Evgeni Gavrilovi #:
I could not understand one point, theSavitsky-Golei filter itselfis not in MT5, then how new data are obtained?

It is enough to feed data to the side for calculation.

[Deleted]  
Evgeni Gavrilovi #:
I could not understand one point, theSavitsky-Golei filter itselfis not available in MT5, then how new data is obtained?
It is used only for marking trades for the training dataset. After markup it is no longer needed, because the model has already learnt when to open and close trades.
 
fxsaber #:

It is sufficient to submit the data for calculation to the party.

Maxim used to have models that didn't require the python script to be enabled

 
Maxim Dmitrievsky #:
It is used only for marking transactions for the training dataset. It is not needed after markup.

I see.

 
in tester mode "all ticks" or "by real ticks"?
[Deleted]  
And #:
in the tester mode "all ticks" or "by real ticks"?
For rough testing you can use opening prices, because the Expert Advisor has control over the opening of a new bar. Then you can use real ticks. The results will not differ much, if the stops are not very short.
 
Maxim, tell us more about how you cluster the price trajectory of a currency pair by the asymmetry coefficient.


12345678910
New comment