Machine learning in trading: theory, models, practice and algo-trading - page 3238

[Aleksey Nikolayev]

Andrey Dik #:
neuro, of course).

Only the final trained grid can be put into the onnx file. It is also possible (theoretically) to put the preprocessing of features for the grid from the initial price. All together it is called a pipeline and the question is how flexible is the possibility to stuff all your pipeline into one ONNX file.

No one really wants to dive into this topic, because you need to be very good at python and there are very few manuals on the topic, and the AI lies as usual).

[mytarmailS]

Aleksey Nikolayev #:

Only the final trained grid can be put into onnx file. It is also possible (theoretically) to put in preprocessing of signs for the grid from the initial price. All together it is called a pipeline and the question is how flexible is the possibility to stuff all your pipeline into one ONNX file.

No one really wants to dive into this topic, because you need to be very good at python and there are very few manuals on the topic, and the AI lies as usual).

preprocessing is roughly speaking normalisation, so the number of features does not change....

I'm wondering if it is possible to generate new attributes inside, the model accepts an array from ohlc, and inside the ohlc format new attributes are already generated, there is still a good piece of full-fledged custom code inside....

I still don't understand whether it can be done with little blood or not...

[Andrey Dik]

it seems that in the future, casts of human consciousness (backups) could be stored in ONNX-like models...

[blef]

We need a template from the organisers, then

a lot of things will become clearer.

[Aleksey Nikolayev]

mytarmailS #:

preprocessing is roughly normalisation, so the number of features doesn't change.

I am wondering if it is possible to generate new attributes inside, the model accepts an array from ohlc, and inside the ohlc format new attributes are already generated, there is still a good piece of full-fledged custom code inside....

I still don't understand whether it can be done with little blood or not...

IMHO, for a Python beginner it is definitely not, for a Python pro it is a question of the limits of possibilities. There are few manuals so far and they contain only simple examples.

[mytarmailS]

Aleksey Nikolayev #:
IMHO, for a Python beginner, definitely not, for a Python pro - it's a question of the limits of possibilities. There are few manuals so far and they contain only simple examples.

Anyway, judging by

http://w ww.xavierdupre.fr/app/mlprodict/helpsphinx/tutorial/onnx_numpy.html

and at

http://www.xavierdupre.fr/app/mlprodict/helpsphinx/api/npy.html#l-numpy-onnxpy

if you just need to transfer the model and pip with simple normalisations, then onnx is good...

but if you need to do something customised, it's either AD or not possible at all...

I don't know what kind of traders this is aimed at...

it looks like 5 people will participate in the contest. 4 of them are metaquotes employees who worked on integrating onnx into metatrader.

[СанСаныч Фоменко]

mytarmailS #:

Anyway, according to

http://w ww.xavierdupre.fr/app/mlprodict/helpsphinx/tutorial/onnx_numpy.html

and by

http://www.xavierdupre.fr/app/mlprodict/helpsphinx/api/npy.html#l-numpy-onnxpy

if you just need to transfer the model and pip with simple normalisations then onnx is good.

but if you need to make something custom, it is either AD or not possible at all....

I don't know what kind of traders this is designed for...

It looks like 5 people will participate in the contest, 4 of them are metaquotes employees who worked on integrating them into metatrader.

I have a real EA with R, with the first version of which I reached the tester.

The structure is as follows:

1. There is an ordinary Expert Advisor on µl, with the usual set of functions: working with orders, stops, MM..... The block of signal generation, in the examples of metaquotes - the intersection of two mashes, is replaced by an appeal to R, which sends the next OHLC.

2. The code of R roughly speaking consists of two parts:

2.1. converting the OHLC into a bunch of predictors for the models. These are hundreds (or thousands) of operators in R with access to some packages (not models) from R.

2.2. actual signal computation by the model.

3. The signal for trading is transmitted back to the Expert Advisor: -1; 0; 1.

Returning to the topic, it turns out that to use ONNX p.2.2 will be ONNX, and in the EA will have to move p.2.1? This is a serious undertaking for me, as besides the models themselves, other packages from R are used, the algorithms of which will have to be coded in µl.

[fxsaber]

Aleksey Nikolayev description of the product to add a mention of the phrases

• machine learning
• neural networks
• artificial intelligence
• GPT and AI

Now there is an interesting phenomenon in society - the popularity of something from the scientific world. Therefore, the behaviour has changed: if before the scientific description repulsed ordinary people, now on the contrary - it attracts them. As a consequence, for more sales it makes sense to mention other near-scientific terms like "Fourier and Laplace transforms". And, of course, beautiful pictures/animations on the topic.

Text and pictures are already enough.

• The world's first world championship in trading between artificial intelligences.
• An unperturbed cyborg against the background of charts and panicked stock exchange workers.
• Crisis? - Artificial intelligence is the future.
• A strong expert in machine learning? - Prove/test your competence. Embellish your CV with a world-class prize.
• Etc.

There will definitely be no shortage of viewers. Attracting experts is doubtful, but it happens. The experts will probably say I don't have a TC. Encourage newcomers to the MoD - it's hard to say. What will definitely work is the number of sales in the Market of EAs with AI. I think additional Market sales would pay for the championship within a year. The budget for organising and running the championship is $250K, minimum. Prize money is a small part of the budget.

[mytarmailS]

СанСаныч Фоменко #:

Back to the topic, it turns out that to use ONNX, 2.2.2 will be in ONNX, and 2.1 will have to be moved to the EA? This is a serious undertaking for me, as besides the models themselves, other packages from R are used, the algorithms of which will have to be coded in µl.

As I understand everything is exactly like this

--------------------------------

this technology is not for humans...

the main thing in a model is the attributes, not the model itself...

Professionals have already established the process of creating attributes and data processing, they also know how to connect models to the market, they just don't need ONNX.

And beginners don't even know ONNX.

Who was this created for, I don't know in my heart ...

For a contest? :)

[blef]

I realised it too - ALL the input data preprocessing has to reside

inside the EA