Machine learning in trading: theory, models, practice and algo-trading - page 2949

New comment
 

Is it possible to implement the following scheme in ONNX:

  • receive the next price in the terminal
  • pass it to the model
  • train the model. Obviously, the model must be in a computing environment in which it can be trained.
  • the model makes a forecast for the next step
  • the received forecast is transferred to the terminal
 
СанСаныч Фоменко #:

Is it possible to implement the following scheme in ONNX:

  • receive the next price in the terminal
  • pass it to the model
  • train the model. Obviously, the model must be in a computing environment in which it can be trained.
  • the model makes a forecast for the next step
  • the received forecast is passed to the terminal
You can't, we've already discussed this.
ONNX is a once trained unchangeable model.
 
СанСаныч Фоменко #:

No one in the world studies an encyclopaedia, they study a specific article. I have given links to a very concrete article. Only get not only an answer to your theoretical question, but also a working code.

Your concrete article is about R and package, actually a manual for software. Completely unspecific, without formulas and disproportionate.

I need to understand the internal logic (one single nuance of the calculations, all others are transparent). The question was addressed to those who may be familiar with the formula (know it and can explain it in two sentences). It was assumed that those who don't know would simply remain silent. Don't slip lengthy instructions with fancy dependencies or sources instead of an answer.

 
mytarmailS #:
You can't, it's already been discussed.
ONNX is a once trained immutable model

You can:

  1. Periodically (once an hour, a day, etc.) you transfer data to a third-party system for additional training
  2. The third-party system retrains and puts a new *.onnx file in the catalogue available to the MQL5 robot.
  3. The robot checks if the *.onnx file has been changed or unloads the old model and loads the new one according to the schedule.
  4. The robot works on the retrained model without interruption

If we're talking about a black-box prediction made on the side, it's not about ML or models at all. It's just a matter of getting a signal from the side.
 
Stanislav Korotky #:

Your specific article is about R and the package, actually a software manual. Completely unspecific, without formulas and disproportionate.

I need to understand the internal logic (one single nuance of the calculations, all others are transparent). The question was addressed to those who may be familiar with the formula (know it and can explain it in two sentences). It was assumed that those who don't know would simply remain silent. Don't slip lengthy instructions with fancy dependencies or sources instead of an answer.

Personally for you I am reattaching exhaustive text on formulas in PDF file. This includes "dependencies and sources".

And about the nuances of calculations, I do not do it, because I know for sure that formulas have NOT anything to do with programming, it is an independent problem, which is solved by other people with other training and in other, scientific circles.

So read the PDF.

Files:
gbm.zip  257 kb
 
Renat Fatkhullin #:

You can:

  1. Periodically (once an hour, a day, etc.) transfer data to a third-party system for additional training
  2. The third-party system retrains and puts a new *.onnx file in the catalogue available to the MQL5 robot.
  3. The robot checks if the *.onnx file has been changed or unloads the old model and loads the new one according to the schedule.
  4. The robot works on the retrained model without interruption.
and such a thing can be put into a Market or a strategy tester?
 
Renat Fatkhullin #:

You can:

  1. Periodically (once an hour, a day, etc.) transfer data to a third-party system for additional training
  2. The third-party system retrains and puts a new *.onnx file in the catalogue available to the MQL5 robot.
  3. The robot checks if the *.onnx file has been changed or unloads the old model and loads the new one according to the schedule.
  4. The robot works on the retrained model without interruption

If we're talking about a black-box prediction made on the side, it's not about ML or models at all. It's just a matter of getting a signal from the side.

If we talk about files, there is a problem when using #property tester_file - if you run a test and after its completion replace the file that should be passed to the Expert Advisor, the latter does not see it - it is solved only by reloading the terminal. The same situation, if you run the test without attaching this file by the link specified in the code, and then put it and run it again, you will get an error due to the lack of the file. It is solved by restarting the terminal. All this is in portable mode on semerka. The problem is many years old - I have written about it many times....

 
mytarmailS #:
and such a thing can be put into a Market or a strategy tester?

I think they have already written that it is possible - what is the difficulty?

 
Renat Fatkhullin #:

You can:

  1. Periodically (once an hour, a day, etc.) transfer data to a third-party system for additional training
  2. The third-party system retrains and puts a new *.onnx file in the catalogue available to the MQL5 robot.
  3. The robot checks if the *.onnx file has been changed or unloads the old model and loads the new one according to the schedule.
  4. The robot works on the retrained model without interruption

If we're talking about a black-box prediction made on the side, it's not about ML or models at all. It's just a matter of getting a signal from the side.

this is exactly the point I would like to solve without " checks if the *.onnx file is modified or scheduled".

How to signal from outside, "hey metatrader - data is ready!". ? How topass an event toan event-driven programme? Formany, many years, NO WAY.

 
Maxim Kuznetsov #:

This is the point I would like to solve without " checks if the *.onnx file is modified or scheduled".

How to signal from the side, "hey metatrader - data is ready!". ? How topass an event toan event-driven programme? Formany, many years, NO WAY.

Yes, the ability to create custom events and their handlers would be very useful.

1...294229432944294529462947294829492950295129522953295429552956...3497
New comment
Reason: