Machine learning in trading: theory, models, practice and algo-trading - page 603

 
Maxim Dmitrievsky:

They say ReLU is better :)

I compared everything in Darche - including ReLU - sigmoid was the best. But the study is fast on the same site. At other sites everything can change.

You have to try and not trust what they say.

Trust but verify.)
 
elibrarius:

I compared everything in Darche - including ReLU - sigmoid was the best. But the study is fast on the same area. On others it may change.

It's necessary to try and not to trust what they say.

Trust - but verify)

And in the process of learning you can include steepness optimization, I actually did so, but only for fuzzy logic. The slope may strongly influence it, yes.

 
Maxim Dmitrievsky:

I still can't even understand what is better - sigmoid or tangent. ) Some people write tangent, now I read that

For most tasks it makes absolutely no difference. It makes a difference when you have some specific requirements for the NS.
 
Yuriy Asaulenko:
For most tasks it makes absolutely no difference. It makes a difference when you have some specific requirements for the NS.

Okay... as soon as I finish the recently suggested book, read some more Winner or something like that... and I will create neural network automata further )

i have only 1 month left, which i allotted for bots with NS :))) i need to keep up the pace... a week for more books, and that's it.

 
Maxim Dmitrievsky:

Okay... as soon as I finish the recently suggested book, read some more Winner or something like that... and I will create neural network automata further )

I have only 1 month left, which I had allotted for myself for bots with NS :))) I need to keep up the pace... a week for the books, and that's it.

This is it, Maxim, you have to work.

You will be ready with a recurrent neural network in 10 lines of code. Use specialized software, and you will be happy.

I showed you before, my network is 3 lines of code.

1. 1. Configure it,

2. Teach it.

3. 3. Work.

Рекурентная нейронная сеть в 10 строчек кода оценила отзывы зрителей нового эпизода “Звездных войн”
Рекурентная нейронная сеть в 10 строчек кода оценила отзывы зрителей нового эпизода “Звездных войн”
  • 2011.01.16
  • habrahabr.ru
Hello, Habr! Недавно мы получили от “Известий” заказ на проведение исследования общественного мнения по поводу фильма «Звёздные войны: Пробуждение Силы», премьера которого состоялась 17 декабря. Для этого мы решили провести анализ тональности российского сегмента Twitter по нескольким релевантным хэштегам. Результата от нас ждали всего через...
 
Yuriy Asaulenko:

That's the way to work, Maxim.

Three days and the network is ready - a recurrent neural network in 10 lines of code. Use specialized software, and you will be happy.

I showed you before, my network is 3 lines of code.

1. 1. Configure it,

2. Teach it.

3. Working.


I have all sorts of them - in 3 and 4 lines :) the very approach of static networks is fundamentally flawed for forex

I'll soon have my own recurrence (a kind of semblance), and there's nothing else to do

 
Maxim Dmitrievsky:

I have all sorts - in 3 and 4 lines :) the very approach of Satellite Networks is fundamentally flawed for forex

there will soon be a recurrence (some semblance), and there's nothing else to do

Is this an axiom from you? )))))

If you train in MT optimizer, then the approach of applying NS to Forex is defective))) IMHO.

 
Yuriy Asaulenko:

Is that an axiom from you? )))))

If you train in MT optimizer, then the very approach of applying NS in Forex is defective.))) IMHO.


yes, i am quick to draw conclusions :)

If you don't realize that you are an optimizer, don't forget... and it will remain an optimizer in static mode...

how do you not understand... i already understood it :)

 
Maxim Dmitrievsky:

Yes, I'm quick to draw conclusions :)

Ns is an optimizer, don't forget... and will remain an optimizer statically

how have you not yet understood... i have already understood :)

Maxim, you forgot, I already have the ns working :)) A simple perceptron, without any frills.

I don't know what to do next - whether to use Python, or R, or even forget about it).

 
Yuriy Asaulenko:

Maxim, you forgot, my NS is already working.) A simple perceptron, without any frills.

I'm not sure what to do next - whether to use Python, or R, or even forget about it).


Well, it doesn't work at Forex, does it?)