Machine learning in trading: theory, models, practice and algo-trading - page 2977

[Aleksey Vyazmikin]

Maxim Dmitrievsky #:
You have a specific approach with specific terminology, people are not ready to sacrifice space on their hard drive for such information, without understanding the result. I see the way out is either to make a detailed article to find enthusiasts who will (most likely) start pushing this in the marketplace right away 😀, or someone to pay a little bit as a journeyman. Or a student for a stick of sausage.

In the specific topic, I kind of use the term "quantum cutoff" from off-topic, which simply means the range of predictor values. Why the term came about - originally from a quant table that divides the predictor via some quantisation formula.

It is difficult to write a useful article here - it is not clear what to write that is not described elsewhere. And what to fill it with - comparative tests or logical calculations. Logic must be supported by formulas, which I can't do.

To take an intelligent student - yes, it would be good if he would understand the essence, not just formulas. Do you have experience in finding such a person?

Maxim Dmitrievsky #:

Here is an example of my last article about meta-models even. How many people wrote with suggestions and improvements, ideas? Zero :) and there is a ready-made TC. How many have taken it and are using it? At least several, without feedback.

There is very little target audience for python articles. In this sense, Dmitriy with his series of articles - click in the tester and get the result. On the idea itself - try to do screening not by 0.5, but by 0.35 - i.e. TN where it is classified with high confidence, and preferably on the control to stop the training of the sample.

Maxim Dmitrievsky #:

It's a very difficult question in general, when you do something and then you are afraid to put it in the public domain, so that you don't turn out to be a sucker. But at the same time you want to be helped. Here there are 1000 kites for 1 Mother Teresa, they will just eat you up. This is such a specific field of activity. Then you're on your own.)

If I even post 3000 lines of code, not including auxiliary classes, who will go into it? That's why it seemed that I asked specifically about metrics characterising the sample, with the specification - within the quantum segment of one predictor, but we've jumped again to the discussion of quantum tables and what good they do.

[Maxim Dmitrievsky]

Aleksey Vyazmikin #:

To get an intelligent student - yes, it would be good if he or she understands the essence, not just formulas. Any experience in finding such a person?

There is very little target audience for Python articles. In this sense, Dmitriy is more interested with his series of articles - click in the tester and get the result. On the idea itself - try to do screening not by 0.5, but by 0.35 - i.e. TN where it is classified with high confidence, and preferably on a control to stop the sample learning.

The student should be able to find you on their own :)

On the article - yes, in general the smaller the fraction of bad trades, the more rare and accurate the trades. Still depends on traits and other tricks. The most important is the principle that an artificial trader can look for inefficiencies on its own. Analogous to how a human can do it. And the more areas of knowledge are used in the preparation of TS, the more neural networks can be in the algorithm, each doing something different. It doesn't have to look like the article.

Dmitry has done a great job, but he won't be able to maintain it all because of the volume. Now there is no point, since there is onnx. What's the usability there? Write tonnes of code to understand something unknown. For example, that RL is an optimisation method like the others. And trader when :)

Let's say you are a trader, but not a mathematician or a programmer. And you start doing scary things: learning programming and formulas... years go by..... Then you find out in the end that even mathematicians couldn't cope with the tasks you set, and you're just some junior. The unenviable fate of a lone trader.

[mytarmailS]

Who's Dimitri?

[Renat Fatkhullin]

Preview of ONNX models directly in the editor is now open:

[Maxim Dmitrievsky]

mytarmailS #:
Who is Dimitri?

RL articles

[Aleksey Nikolayev]

Renat Fatkhullin #:

Preview of ONNX models directly in the editor is now open:

It seems to be common to visualise them with graphs.

What in principle can visualisation do?

[Aleksey Vyazmikin]

Maxim Dmitrievsky #:
The student has to find you himself :)

Hmm, how would he find out more :)

Maxim Dmitrievsky #:

According to the article - yes, in general, the smaller the fraction of bad trades, the more rare and accurate trades are.

I just wrote about smaller volume of correctly classified bad examples due to their classification accuracy.

Maxim Dmitrievsky #:

Dmitriy has done a great job, but he will not be able to maintain it all because of the volume. Now there's no point, because there is onnx. What kind of usability is there? Write tonnes of code to understand something unknown. For example, that RL is an optimisation method like the others. And when to trade :)

He has very hard to read code there, but you can figure it out in general. I think it is a very good incentive for his personal understanding of MO. And reproducible code is very important for understanding the process. Especially if someone wants to make something of his own.

Maxim Dmitrievsky #:

Let's say you are a trader, but not a mathematician or a programmer. And you start doing scary things: learning programming and formulas... years go by.... Then you find out in the end that even mathematicians couldn't cope with the tasks you set, and you're just some junior. The unenviable fate of a lone trader.

Yeah, that's what happens - wasted years.

[Aleksey Vyazmikin]

mytarmailS #:
Who's Dimitri?

Here are his articles.

[Aleksey Vyazmikin]

I've put my problem in a separate thread - anyone who wants to help is welcome!

[Maxim Dmitrievsky]

Aleksey Vyazmikin #:

Hmmm, how would he know more of that :)

I just wrote about the lower volume of correctly classified bad examples due to their classification accuracy.

His code there is very hard to read, but you can figure it out in general. I think for his personal understanding of MO this is a very good incentive. And reproducible code is very important for understanding the process. Especially if you want to make something of your own.

Yes, that's what happens - wasted years.

Well, so far there's nothing to discuss. It's all in Sutton and Barto's book and then dip RL. But the volumes are amazing, these guys are working at uni for a good wage. What a head it should be, it would be in the right direction :) Interested in exhausted approaches, based on experience of application.

As they said, MO in trading has already moved from the underground to the mainstream, but still at the level of tester grails. And indicator fever is over :)