"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 19

New comment
 
TheXpert:
By the way, just a side note. Most packages don't have genetics for networks at all. And in this list, it's only for checkboxes or if heuristics are missing (imho).
By the way, it's worth mentioning. I tried to use this software in demo-version. For the test screwed up such a scheme that the devil himself would break a leg - best coped with the task is the GA. The rest either start to stumble, or get stuck in the locals. Therefore, GA there and included in the composition - for particularly severe cases. :)
 
Urain:

By the way, why don't they have BeckProp? Or did I misunderstand something?

Yes, they don't. Apparently it's not there because it's problematic to use with large schemes of miscellaneous networks and pre/post handlers.
 
joo:
Yeah, we don't have it. Apparently it's not there because it's problematic to use with large schemes of heterogeneous networks and pre/post handlers.

So maybe we should use these three paradigms as a basis?

Genetic algorithms





Particle swarm





Monte carlo


Or is there something else to add?

 
Urain:
So why don't we use these three paradigms as our basis?
Yeah, go for it. I'll just wash my hands of this and start an alternative project:) .
 
gpwr:
By the way, Vladimir, would you like to voice your view and grids more broadly?
 
Urain:

So maybe we should use these three paradigms as a basis?

Genetic algorithms





Particle swarm





Monte carlo


Or should I add something else?

The point is that these algorithms are universal and can cling to anything.

There are more, but not very popular "universal", but we don't need them:

Линейное программирование — Википедия
  • ru.wikipedia.org
Линейное программирование — математическая дисциплина, посвящённая теории и методам решения экстремальных задач на множествах -мерного векторного пространства, задаваемых системами линейных уравнений и неравенств. Многие свойства задач линейного программирования можно интерпретировать также как свойства многогранников и таким образом...
 
TheXpert:
Uh-huh, take it. I immediately wash my hands and start an alternative project :) .

The main thing with a pure heart :o)

Andrey continue to do what you're doing, I (as will become freer) will engage in graphical input engine, well, in parallel, mine, and there time will show what is better, then people will catch up.

 
TheXpert:
Uh-huh, take it. I immediately wash my hands and start an alternative project :) .
It's always like this with you smart guys. No compromise. "Either do it my way or wash my hands of it." You can't do that.
 
joo:

The point is that these algorithms are universal, and cling to whatever you want.

There are more, but not very popular "universal" ones:

For example, Newton's method is defined only for known function, if you don't know function type you can't calculate it outright, quasi-Newtonian methods are already used for that (I can't say about others, but I guess they have their limitations too).

Again it is worth to apply evolutionary sifting model, if method is unknown then for sure it is lame (unless it was invented yesterday and is just poorly known). Most optimization methods are about 300 years old.

 
Mischek:
It's always like this with you smart guys. No compromise. "Either do it my way or wash your hands of it." You can't do that.
The fundamental point here is that all of these methods require additional memory for learning.
1...121314151617181920212223242526...100
New comment