Machine learning in trading: theory, models, practice and algo-trading - page 2765

 
Maxim Dmitrievsky #:

I don't remember what it is ) maybe from an article on seasonals

Anyway, I don't touch time now, if the analogy with the tripping window is true, it's not about time as such.

but in general yes, the lag of increments will change and the window itself will shift. It can be done in different ways, I haven't done something specific yet.
Got it, time in increments)))) and search for the whole row. Thanks)
 
JeeyCi #:

Do you have any works (products, articles) on this topic? Can I take a look?

How, in your opinion (Vladimir Perevenko refused to answer this question for me))? - is it possible to make relatively stable money on neural networks in trading?

 
Valeriy Yastremskiy #:
Got it, time in increments)))) and search the entire row. Thanks))

I wanted to do it like in the last article, but the second NS will not filter bad examples, but will be responsible for the window.

although timing is also important, maybe add a 3rd NS to the algorithm, and all this will be fun to optimise.

there are still a few promising algorithms on hold (according to my slippers)

 
Maxim Dmitrievsky #:

I wanted to do it like in the last article,but the second NS will not filter bad examples, but will be responsible for the window.

although timing is also important, maybe a 3rd NS could be added to the algorithm, and all this would be fun to optimise.

there are still a few promising algorithms on hold (according to my slippers)

Very curious!

Is this with a teacher? If with a teacher, what is the teacher and what are the predictors?

 
СанСаныч Фоменко #:

Very curious!

Is it with a teacher? If with a teacher, what is the teacher and what are the predictors?

With a teacher, the first NS works on buy/sell, the second one adjusts the window shift based on the results of the first one's predictions. For example, at each new bar it tells whether to shift the window with signs or leave it at the previous bar. The bundle of 2 NSs is retrained several times, improving the results. Predictors are not crucial yet, you can add any of them you want

Well, this is purely creative, we can finalise the scheme based on the results.
 
Maxim Dmitrievsky #:
With the teacher, the first NS works on buy/sell, the second one adjusts the window shift according to the results of the first one's predictions. For example, at each new bar it tells whether to shift the window with signs or leave it at the previous bar. The NS binder is retrained several times, improving the results. Predictors are not crucial yet, you can put in any of them you like

What's all the fuss about?

Do you have a graph "Prediction error - window size"? Or any other similar information?

 

new section of the special Olympiad: one NN trades on ticks, the second predicts the results of the first one, and the third one warns of the second one's fakups. (it can be combined indefinitely). The battle is fought on bets and forward, because even on demo it's a bit scary to run this game.

 
Ivan Butko #:

Do you have any works (products, articles) on this topic? Can I have a look?

How, in your opinion (Vladimir Perevenko refused to answer this question for me))? - is it possible to make relatively stable money on neural networks in trading?

the question of earnings and the fact of using or not using neural networks have no linear or any other dependence, except for a very strongly indirect one - any HR will tell you that and even an employee.... - you can look up anything on the internet... if I don't use this thread as an advertising platform, then it is not intended for that purpose... I'm not here to promote brands/products/etc, -- the topic is only interesting where there is a reasonable justification... == if you can't make a review of the information available in various sources on the issue you are interested in, whether it's market analysis or something, I don't even know what it is and why, it's probably too early for you to ask for sources ... )) you don't know how to use citation in your work yet....
 
СанСаныч Фоменко #:

What's all the fuss about?

Do you not have a "prediction error - window size" graph? Or any other similar information?

By my calculations, the window size ranges from 700 to 2000 bar. Predicting one bar ahead takes 35 seconds. If a blunt overshoot, i.e. increase the window from 700 to 1800 and select the optimal one, we get 420 seconds This is extremely long. Lots of things to optimise a lot, but requires other computers and the ability to load all cores.

 
JeeyCi #:
the question of earnings and the fact of using or not using neural networks have no linear or any other dependence, except for a very strongly mediated one -- any HR will tell you that and even an employee.... - you can look up anything on the internet... if I don't use this thread as an advertising platform, then it is not intended for that purpose... I'm not here to promote brands/products/etc, -- the topic is only interesting where there is a reasonable justification... == if you can't make a review of the information available in various sources on the issue you are interested in, whether it's market analysis or something, I don't even know what it is and why, it's probably too early for you to ask for sources ... )) you don't know how to use citation in your work yet....

I understand you, thank you for your reply.