Machine learning in trading: theory, models, practice and algo-trading - page 1331

New comment
 
elibrarius:

Reshetov's nuclear machine is the same bicycle that some here use. And it seems to be more successful in the market than something standard.

So I'm for bicycles! ) But of course you should understand what to do with them too.

It's a bicycle for you and a kernelized SVM for others.

Again, you don't know what you're writing about but you do... here's your cyclist approach

instead of reading a couple of reference books to sit around and make up nonsense... that's the approach
 
Maxim Dmitrievsky:

The problem is not in the standard methods, but in the elementary lack of understanding of what you are trying to do with them and what process you are working with

i.e. the lack of both economic and mathematical education

so your actions are like a Brownian particle wandering... maybe this way, maybe that way...

And everyone refuses to read "complicated" books, especially in English.

So tell me, where am I wrong, what are my conclusions. Otherwise, by evaluating your actions without pointing out the error you just show your merits, but everyone knows about them anyway...

And I'm an economist by education, so don't make such far-reaching conclusions.

 
Elibrarius:
You have to make up bicycles during the day. And sleep at night. You should take care of your health.

I can't do it during the day...

Thank you for your concern.
 
Aleksey Vyazmikin:

So explain what I am wrong about, what conclusions I draw. Otherwise, giving an assessment of actions without pointing out the error, you just demonstrate your merits, but everyone knows about them anyway...

And, I am an economist by education, so do not make such far-reaching conclusions.

You're an economist, but maybe you're not familiar with econometrics.

I already wrote that you're trying to manage random behavior by changing the seed and the number of elements of sub-samples, when there are described methods of working with samples (described in books that everyone hates), i.e. you are profaning the MO

I won't cite the literature for the 100th time, maybe there are better books, especially since I mostly read in English
 
Maxim Dmitrievsky:

This is a bicycle for you and a kernelized SVM for others

again, you don't know what you're writing about but you do ... here's your cyclist approach

instead of reading a couple of handbooks to sit and make up crap... that's your approach

I know superficially - no arguments.
But Reshetov also did not use something alien, and decided to create their own, with their own features and tricks. At the initial stage of development it was exactly a bicycle.

It is out of making up nonsense that sometimes successful ideas are found.

 
elibrarius:

I know it superficially - no arguments.
But Reshetov, too, did not use something foreign, but decided to create his own, with his own features and tricks. At the initial stage of development it was exactly a bicycle.

It's out of making up nonsense that sometimes good ideas are found.

Once again: he didn't invent anything, but took a ready-made machine learning method, the name of which is written above

Developed by Vapnik, partly by Ivakhnenko, but not by Reshetov.

and he came up with idiotic terminology that takes away from the essence of the method
 
Maxim Dmitrievsky:

You are an economist, but maybe you are not familiar with the theory of econometrics (at least)

I have already written that you are trying to manage random behavior by changing the seed and the number of elements of sub-samples, when there are described methods of working with samples (described in the books hated by everyone), i.e. you are profaning the MO

Yesterday you said that methods from books very bad when applied to BP, but now the contrary, you refer to the literature. You know that most books describe stationary processes, or cyclic time series, and what we do not describe any book I know (if you know such - name it, please), so any experiments lead to an understanding of the process, and can lead to new ideas.

Seed arranges random - what's wrong with that, if the random can be, conventionally speaking, over sampled?

Let's talk specifically about sampling, what proportion should we stick to, based on the literature?

 
Aleksey Vyazmikin:

Let's be specific about sampling, what proportion should be followed, based on the literature?

Seed arranges the random - what is wrong with that, if the random can be, conditionally speaking, over-sampled?

find yourself about sampling and read, Ivakhnenko "MSUA" - good books in Russian

This is another made-up nonsense about random seed ordering. Are you building a model or looking at random events?

Show me where it says that ME models are built on seed search

 
Maxim Dmitrievsky:

find yourself about sampling and read, Ivakhnenko "MSUA" - good books in Russian

The fact that seed orders Random is just another load of crap that has been made up. Are you trying to build a model or are you just looking at random events?

You don't want to talk about sampling, fine. Thanks for the offer to read specific literature.

But I do not understand your logic about Seed - each time when you start creating a model a random variable is generated that can be fixed using Seed parameter, this variable affects the creation of the model, so saying "are you building a model or looking through random events" brings me to a logical stupor - explain your thought, please.

Maxim Dmitrievsky:

Show me where it says that MO models are built on brute force seed

Models are built on predictor relationships, Seed affects the formation of those relationships and thus model building. What is the contradiction - I can not understand!?

In addition, in the lectures they recommend to brute-force this Seed, including yesterday's video on CatBoost seed in the example file in python is fixed - clearly obtained by brute-force.
 

live-forex-signals.com

1...132413251326132713281329133013311332133313341335133613371338...3623
New comment