What to feed to the input of the neural network? Your ideas... - page 12

New comment
 
The input should be an array of structures that represents a hierarchy of recognised patterns across the history.

For example, parabolic channels:


As a rule, there are 7-10 nested channel objects in such a hierarchy.
 
Nikolai Semko #:
The input should be an array of structures that represents a hierarchy of recognised patterns across the history.
For example, parabolic channels:
Something of a higher order.
Thank you for your reply, I am not comfortable with OOP terms yet
If you have experience, I would be glad to have a couple of lines of opinion, what kind of results this kind of data gives and on what machine(model)
 
Ivan Butko #:
Something of a higher order.
Thank you for your reply, I am not comfortable with OOP terms yet
If you have experience, I would be glad to have a couple of lines of opinion, what results this kind of data gives and on what machine(model)

yes, that's right, it's from higher order, but there's no other way to do it.
In lower orders, based just on quotes is just mouse fiddling. Neuronka will constantly build something like Fourier extrapolation, which has zero forecasting efficiency.
If you feed MO with a similar structure on each tick throughout history with the price behaviour in the future after each current structure, but here the real creativity begins, which gives very good forecasting efficiency.
Without OOP is also impossible. To master OOP is 2 hours of theory and 50 hours of practice.

 
Nikolai Semko #:
The input should be an array of structures that represents a hierarchy of recognised patterns across the history.

For example, parabolic channels:


There are usually 7-10 nested channel objects in such a hierarchy.
Thanks for the informative post. How many images do you have approximately? How is the hierarchy formed?
 
There are a lot of topics about machine learning, but for some reason everywhere it is about price prediction. This approach only discredits the MOE. Remember from the TV series: "to find rails, you have to think like rails".
 
G1G2G3 #:
Thanks for the informative post. How many images do you have approximately? How is the hierarchy formed?

I just gave an example based on objects - parabolic channels. The hierarchy of objects or even processes can be anything.
I just specialise in recognising exactly channels (linear, parabolic or wavy, i.e. polynomial of degree 1,2 or 3). Been doing this for about 12 years now. I published the first step in this direction here. Current algorithms are thousands of times faster than in this code and of higher quality.
Hierarchy is found with the help of recognition algorithms by its long history.

 
Sergey Pavlov #:
There are a lot of topics about machine learning, but for some reason everywhere it is about price prediction. This approach only discredits the MoD. Remember from the TV series: "to find rails, you have to think like rails".

And simply the sense of MO is lost if it is not used to open a position at a more favourable price and the corresponding subsequent sale.

But, if anything, this topic is not about MO. It's about trying to exploit any parts of that broad topic)

 
Ivan Butko #:
But, if anything, this thread is not about the MoD

I'm curious, what does it concern?

 
Sergey Pavlov #:

I'm curious, what does it address?

What to feed into the neural network's input
 
Sergey Pavlov #:
There are a lot of topics about machine learning, but for some reason everywhere it is about price prediction. This approach only discredits the MoD. Remember from the TV series: "to find the rails, you have to think like the rails".
It's just a question of terminology.
I don't like the term forecast either.
But such a term is more understandable for others than, for example, probability clouds.
Why the deceit?
After all, every time a trader or EA, opening a trade, makes a forecast that the price will move in this direction with a probability of more than 50% based on some data.
Probabilistic prediction is the ultimate goal of trading. Fact. MO is just a tool for that purpose.
1...5678910111213141516171819...67
New comment