Machine learning in trading: theory, models, practice and algo-trading - page 1525

[Vladimir Simakov]

elibrarius:

7-8 minute commentary on our topic from the practitioner

IMHO of course, but does not give the impression of madam kfmn, and broadcast format a la Malakhov and Co.

[Александр]

Petros Shatakhtsyan:

If they were discussing automatic self-optimization, that would be more interesting than machine learning.

Or no one raises the question that a self-learning robot where supposed to keep all its knowledge and how it will quickly search for it. Not saying that it has to analyze it and learn from it again when a new tick comes along.

I.e. it is a never-ending process, and the robot will not be able to catch up with time and will always fall apart every time in extraordinary situations, not knowing what to do.

Why isn't machine learning an optimization task? it's all the same thing...

[Mihail Marchukajtes]

Alexander:

Why isn't machine learning an optimization task? It's all the same thing...

Optimization is present both in ordinary EAs and in ones with NS, but in simple ones you are limited by the range of optimized parameters, and in NS by the range of weight coefficients, which is a thousand orders of magnitude greater than the range of any optimized ordinary parameters with the same number of them. Did I answer your question?

[Грааль]

Alexander:

Why isn't machine learning an optimization task? It's all the same thing...

Well, there is no clear boundary, as well as between classical statistics and MO. But in general, by "optimization", we mean numerical methods of finding different conditional points (extrema etc.), usually these are iterative approximate methods and in MO there are enough non-iterative and non-closer algorithms, for example linear regression. Some MO algorithms are trained by optimization methods, "annealing" etc. But it is probably not correct to reduce MO to optimization at all.

[Andrey Khatimlianskii]

What do many respected minds have to say about the article?

I stumbled across it by accident, I liked the style of presentation. How much of the content is true?

[Roman]

elibrarius:

7-8 minute commentary on our topic from a practitioner

The 82nd year of the USSR

[Roman]

Andrey Khatimlianskii:

What do many respected minds have to say about the article?

I stumbled across it by accident, I liked the style of presentation. How much does the content correspond to reality?

Very cool article, thanks for finding it.
The article reveals very well what's what, with detailed explanations and diagrams.
I started to read the article and already understood in what direction I need to study the material for my long-standing desire.
On the question which I have asked here, and did not get an answer, and in the article it found.
I haven't finished reading the article yet, but I feel that I need to reread it several times, at different periods of time, in order to catch the necessary details.
And save the html page to my archive, just in case.
To your question, "How much of the content is true?"
This is basically a framework of understanding, which is written in very clear language.
And it seems to me that the local minds that are trying to use the decision trees, go the wrong way.
Because it was originally chosen the wrong type for implementation, as it is clear from the article decision trees belong to the type of classification.
This type is designed to predict the category of the object, not the prediction of the numbers.
Perhaps I'm wrong, as I can not know the task at hand.
Again, this is a basic understanding of where to start, and as the article says, there are many varieties of types to solve problems.
The main thing is to choose the right type for your task.
Something like that, went on reading))

[Roman]

Recently, I needed to contact the support of the mobile operator to clarify my questions.
Naturally, in order to get a quick answer, I chose an online chat on the cellular company's site, hoping that there would be a live person on the other end.
But after a couple of questions and getting an answer, I immediately realized that I was being answered by a chat-bot, because the question I asked him could not answer correctly,
and kept asking me to clarify the question, please clarify the question.
Once again, getting such an answer, I wrote him that you are a stupid and useless bot.
(To which he honestly told me, sorry, I was just learning)).

Just remembered this moment of communication with the bot)

[Forester]

Roman:

Since you initially selected the wrong type to implement, as it is clear from the article decision trees refer to the type of classification.

Not only. They can do regression, too.

[Roman]

elibrarius:

Not only that. They can regress, too.

Okay, I get it. So regression on trees extends the capabilities of conventional regression?
And another question, regression and gradient descent, are they similar algorithms to solve the problem or different?
If they are similar, which algorithm is more accurate?