Machine learning in trading: theory, models, practice and algo-trading - page 2934

[Aleksey Nikolayev]

Renat Fatkhullin #:

A beta version of 3584 with ONNX support is available.

Need Windows 10 build 1809 and newer, preferably Windows 11 with all updates.

What versions of Windows Server are suitable for ONNX?

[Vladimir Perervenko]

Aleksey Nikolayev #:

And which versions of Windows Server are suitable for ONNX?

Have a look here and here

First you need to decide

• why you need it
• who needs it.

It won't get into the market, unless there's a "trained DNN models" section. It is unlikely to work in a tester. Traders should be seriously motivated to master this difficult tool. It is intended primarily for the implementation stage. And many people have solved this stage of MO (good or bad - another question).

Therefore, we are waiting for a detailed description from the developers.

[Aleksey Nikolayev]

Vladimir Perervenko #:

Look here and here

First you have to make up your mind:

• why it's needed
• who needs it

It will not be allowed in the market, unless there will be a section "trained DNN models". It is unlikely to work in the tester. Traders should be seriously motivated to master this difficult tool. It is intended primarily for the implementation stage. And many people have solved this stage of MO (whether good or bad is another question).

I didn't find about server versions on your links. They write here that a 2019 server is needed.

I am interested in universal, costless and fast (in all senses) possibility to run EAs with MO models on VPS. Well and still hope to work in the tester. Possible problems with the marketplace are not important (personally to me).

Vladimir Perervenko #:

So we are waiting for a detailed description from the developers.

Yes, very much waiting.

[Valeriy Yastremskiy]

Aleksey Vyazmikin #:

I understand that 1.13.1 is the latest version for windows 7. So it won't work? The joys and sorrows of a poor MT5 user....

Does it require an additional library to work?

Articles are good - we will wait.

Bought 10k 3 pcs from geekom with celeron 5095 8 gb ozu and 128 hardware expanded to 237 gb in the office and with 11 pro. Apparently one will have to take it for myself, or 11 pro to fill up)))) with drivers I hope to cope with__

[Vladimir Perervenko]

Valeriy Yastremskiy #:

Bought 10k 3 pcs from geekom with celeron 5095 8 gb ozu and 128 hardware expanded to 237 gb in the office and with 11 pro. Apparently one will have to take one for myself, or 11 pro to fill)))) with drivers I hope to cope with__

What is this about so telegraphically?

[Renat Fatkhullin]

Vladimir Perervenko #:

It will not be allowed in the market, unless there will be a section "trained DNN models". It is unlikely to work in the tester. Traders should be seriously motivated to master this difficult tool. It is intended primarily for the implementation stage. And many people have solved this stage of MO (whether good or bad is another question).

Therefore, we are waiting for a detailed description from the developers.

It will be available in the market, in the tester too, including Cloud Network.

ONNX Runtume will be rewritten and redesigned to not depend on obsolete system libraries.

ONNX files are easily included in projects, encrypted and compressed inside EX5 files. You get clean robots from a single file.

Workflow:

1. Train the model on the side, e.g. in Python
2. Convert it to onnx
   
3. Incorporate it into a robot and use it (trayning is not available, only inference).
   

[Valeriy Yastremskiy]

Vladimir Perervenko #:

What's that so telegraphic about?

About the 11pro, cheap and cheerful. You can get a copy off the disc. Thin client hardware at its best. Just lucky, and I'm allergic from 7 to 10)

[Valeriy Yastremskiy]

Renat Fatkhullin #:

It will be available in the mart, in the tester too, including Cloud Network.

ONNX Runtume will be rewritten and redesigned to not depend on outdated system libraries.

ONNX files are easily included in projects, encrypted and compressed inside EX5 files. You get clean robots from a single file.

Schema of operation:

1. Train the model on the side, for example in Python
2. Convert it to onnx
   
3. Incorporate it into the robot and use it (trayning is not available, only inference)
   

I apologise for a possibly stupid question, are ticks in python training available?

[Renat Fatkhullin]

Valeriy Yastremskiy #:

I apologise for the possibly stupid question, are tics in python training available?

https://www.mql5.com/ru/docs/python_metatrader5/mt5copyticksfrom_py

Python training is done outside the tester by itself. Please read the documentation section about Python.

[Aleksey Nikolayev]

I took the model with the lowest version from here, put it in a common folder, ran the script:

void OnStart()
  {
   string mn = "mobilenetv2-7.onnx";
   long h = OnnxCreate(mn, FILE_COMMON);
   Print(h != INVALID_HANDLE);                              // false
   Print(GetLastError());                                   // 5801
   ResetLastError();
   int f = FileOpen(mn, FILE_COMMON|FILE_BIN|FILE_READ);
   Print(f != INVALID_HANDLE);                              // true
   Print(GetLastError());                                   // 0
  }

No description of error 5801 found. My system: Windows 10 build 19045. According to the table should work.