"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 52

 
Urain:

Oh you gynecologist naturals :)


Well, I mean it in a good way ))
 
Mischek:
Well, I mean it in a good way.)
And I mean that in a researcher way.
 

Andrew TheXpert

so wait for your response today to

Urain:
...

I'd rather have your thoughts on the crap I've been spouting here,

or an alternative combat plan.

or can we get out of here?

 
Urain:

Grid engine
1. Initialization of the grid
2. Working stroke of the grid
3. Grid training

That's it? And the working stroke of the grid is what?

And we don't say a word about the tester. Are we going to use the regular one? If homemade, what type? - I imagine there are only two types of tester - with a peek into the future (it has its pros and cons, easier to implement) and without a peek (also has pros and cons, more difficult to implement but more versatile).

And, there seems to be something else missing.

 
Urain:or can I fall off?

Fall off, I'm just accepting.

Urain:

Andrei, agree that the neuro-engine is not a very mastable project.

Nikolai, "not very" and "not" are different things. A project for a couple of months of hard work by one person can already be called small.

 
joo:

That's it? And the working stroke of the grid is what?

And we don't say a word about the tester. Are we going to use the regular one? If homemade, what type? - I can imagine there are only two types of tester - with a peek into the future (it has its pros and cons, easier to implement) and without a peek (also has pros and cons, more difficult to implement but more versatile).

And, I guess there's something else missing.

For engine, solving these three tasks will be enough, the rest of the meat, such as preprocessing or postprocessing is external to the engine methods.

The workflow is the calculation of the signal passing through the network. The very thing that people call the black box.

GA or other method of learning have internal methods, external they only have external error calculation, for GA it is FF.

 
joo:

That's it? And the working stroke of the grid is what?

And we don't say a word about the tester. Are we going to use the regular one? If homemade, what type? - I can imagine there are only two types of tester - with a peek into the future (it has its pros and cons, easier to implement) and without a peek (also has pros and cons, more difficult to implement but more versatile).

And, there seems to be something else missing.

Regular would not be desirable. I would like to use the network not on the quotes history of MT5 tools.

I would like, well personally, to use before financial instruments, for example an hourly chart of temperature in St. Petersburg in the presence of data on the pressure and wind strength, and would start quite simply -

add known distortions to the sine wave and see how they would be caught. With what and how you can catch them. Where it will start to go dull, if it starts to go dull. I want to first get acquainted with the possibilities of neuro on something simple, known to be understood.

 
Urain:

For the engine, solving these three tasks will be enough, the rest of the meat, such as pre-processing or post-processing is already external to the engine methods.

The workflow is the calculation of the signal passing through the network. The very thing that people call the black box.

GA or other method of training are internal methods, external they have only the calculation of the external error, for GA is FF.

I mean, is it possible to develop its parts without a schematic description of the whole machine? At least there should be a ToR (from the engineer's point of view ToR - description of main characteristics of the machine, overall dimensions, area of applicability, cost, competitiveness, aesthetics....). Without this, no chief designer will start designing a machine.

An engineer's ToR does not mean the same thing as a programmer's ToR.

 
joo:

I mean, is it possible to develop its parts without a schematic description of the whole machine? At least there should be a ToR (from the engineer's point of view, the ToR describes the main characteristics of the machine, overall dimensions, area of applicability, cost, competitiveness, aesthetics....). Without this, no chief designer will start designing a machine.

The engineers' TOR is not the same as the programmers' TOR.

How do you.

1. initializing the mesh
2. working stroke of the mesh
3. training the mesh

not a schematic description?

Mischek:

I wouldn't like the in-house one. I would like to use the grid not on the price history of mt5 instruments.

I personally would like to use before financial instruments, for example hourly chart of temperature in St. Petersburg with data on pressure and wind strength.

add known distortions to the sine wave and see how they would be caught. With what and how you can catch them. Where it will start to go dull, if it starts to go dull. I would like first to get acquainted with neural capabilities on something simple, known to be understandable.

That's why I'm trying to separate the grid from the data.

 
Urain:

GA or another method of learning there are internal methods, external they have only the calculation of the external error, for GA it is FF.

No. GA is external because it doesn't need to know the insides.