Machine learning in trading: theory, models, practice and algo-trading - page 2940

[Vladimir Perervenko]

mytarmailS #:
The usual ones, the same "random forest" "intrees" "arules" "arules sequence" "dbscan".

But that's 5%.

And the other 95% is self-written stuff.

If you can't embed any of your own code in onnx, then I'm not interested in it.

Also, half of my algorithms are rebooted at every step, as I heard this doesn't work with onnx.

Of course not. In a very primitive way, the sequence of obtaining a model in ONNX format is as follows: you create a model, train it, optimise it. Then in a special programme (converter) you pass a unit of input data through the trained model. The converter records the sequence of calculations of the forecast model and saves it in onnx format. This model can be run on any platform that has onnxruntime. Only forecast/predict no truntime.

I don't know any package in R that has a converter to ONNX. Maybe torch(R) package will add it, but it should be requested to the package developers.

Probably there are other conversion possibilities, but in Python. I haven't looked into it in depth. Have a look here

[Renat Fatkhullin]

https://learn.microsoft.com/ru-ru/windows/ai/windows-ml/onnxmltools

ONNXMLTools allows you to convert models from various machine learning toolkits into ONNX format.

Installation and usage instructions are available in the ONNXMLTools repository on GitHub.

Support

The following toolkits are currently supported:

• Keras ( keras2onnx con verter shell);
• Tensorflow ( tf2onnx transducer shell);
• scikit-learn ( skl2onnx converter shell);
• Apple Core ML;
• Spark ML (experimental mode);
• LightGBM
• libscm;
• XGBoost;
• H2O
• CatBoost

Pytorch also has a built-in ONNX export facility. See here for more information.

[Renat Fatkhullin]

To simplify code writing in MQL5, we will most likely abandon the current set of functions with handles and switch to the object model.

That is, we will introduce a new built-in onnx object type with convenient methods.

[mytarmailS]

Despite all the mega hype, onnx will be useful only for sellers with very good skills in MO.

And there's a lot of them here. How many of them are there?

[Maxim Kuznetsov]

mytarmailS #:
Кароч несмотря на весь мега хайп, onnx будет полезна только продавцам с оч. Хорошим умением в МО.

and you also say that inside the robot onnx , deep learning, non-linear filters and astral streams...nobody knows that there are 3 mashka's at the core :-)

[Aleksey Nikolayev]

It seems that R is not doing well with ONNX support. I found only onnx package - an interface to this format. This package is based on python and does not seem to be developed. I have not found any analogue of ONNXMLTools for R. Sad.

It's time to learn python. And return to linux - I don't want to contact VS at all.

[mytarmailS]

Aleksey Nikolayev #:

It seems that R is not doing well with ONNX support. I found only onnx package - an interface to this format. This package is based on python and does not seem to be developed. I have not found any analogue of ONNXMLTools for R. Sad.

It's time to learn python. And go back to linux - I don't want to mess with VS at all.

Do you have a working MO model???

If you have a model, are you going to sell it???

If one thing is not true, then the question is, why study it?

[Aleksey Nikolayev]

mytarmailS #:
Do you have a working MO model????

If you have a model, are you going to sell it????

If any of these things are not true, then the question is, why study it?

I've written it before. Interested in:

1) Running TC on MO model in MT5 tester.

2) Running TC on MO on VPS without any additional crutches, quick and easy.

3) Market is NOT interested.

[mytarmailS]

Aleksey Nikolayev #:

Written before. Interested in:

1) Running TC on MO model in MT5 tester.

2) Running TC on MO on VPS without any additional crutches, quick and easy.

3) Market is NOT interested.

Well, you can also test in R

Only VPS is left, and is it worth it?

[Evgeny Dyuka]

Aleksey Nikolayev #:

Written before. Interested in:

1) Running TC on MO model in MT5 tester.

2) Running TC on MO on VPS without any additional crutches, quick and easy.

3) Market is NOT interested.

I think Market is the main thing here.
Everything else is solvable. If only for trading, MO can be connected to MT5 without any problems, but you can't get to the Market that way.
That's why on the scales you can put dancing with tambourine on one bowl and the Market on the other.