What to feed to the input of the neural network? Your ideas... - page 70

New comment
 
Maxim Dmitrievsky #:
The evolution of AI according to the forum. Rosenblatt perceptron -> Reshetov neuron -> Butko neuron 😄

Reshetov is ahead of the curve so far. He even wrote a separate software :)

Well, people don't like to read (even the basics), what can you do.


But they can spend three years training neurons etc... and then you hear questions like "what is the difference between classification and clustering?" and you fall out.

[Deleted]  
Input neural networks with normalized, structured data relevant to your task, like images, text, or numerical data. Preprocess features for consistency, ensuring compatibility with the model architecture and objectives.
[Deleted]  
mytarmailS #:

Well, people don't like to read (even the basics), what can you do.


But they can spend three years training neurons and so on... and then you hear questions like "what is the difference between classification and clustering?" and you fall out.

It is long and difficult to read :)
 
Maxim Dmitrievsky #:
The evolution of AI according to the forum. Rosenblatt perceptron -> Reshetov neuron -> Butko neuron 😄

Reshetov is ahead of the curve so far. He even wrote a separate software :)
Upphpahaha

Neuron Butko - and this is how foreign users will read me in auto-translation

The trace in history is left, we can leave now
 
astrologerdevanand neural networks. Pre-process features for consistency, ensuring compatibility with the model architecture and goals.
You see, it's an applied question here: what and how

After all, a price graph is not an earthquake graph, there is an encrypted language here
 

Where MLP is useful

I keep trying to attach this specific tool.

And I found it:

This thing is unlikely to work "head-on" when you expect entry and exit points from it. BUT!

As a filter it gives much more interesting results. At least, it seems to be even useful.

That is, as an independent intellectual mechanism MLP does not pull in any way, but as an auxiliary tool it can be a bit better than any pomo indicators.

The point is to add the MLP result to the condition ofopening positions. 

if (Бла-бла1 > бла-бла2 && MLP > 0.6)           trade.buy(...)

And in this form MLP can be more interesting than a single indicator or a bunch of several indicators, because it can create some additional signals non-linearly from one of them or from the same bunch.

Yes, the strategy search in this case continues manually (something intelligent manually collected), and MLP will be like a new super-indicator-filter

Optimisation can take two forms:

1) Parameters are optimised first, then MLP is switched on and weights are optimised

2) All together optimised simultaneously


So let it live, it can still be twisted and turned.

 
Ivan Butko #:

So let him live, he can still be twisted and turned.

Let it live. I am preparing material on this topic, with a ready-made Expert Advisor. Even from the data from the simplest indicators MLP can make a candy.

 
Andrey Dik #:

Let it live. I am preparing material on this topic, with a ready-made EA. Even from the data from the simplest indicators MLP can make a candy.

And this is already interesting
 
Looks like the "invention" of ensemble from MO is coming.
 
Aleksey Nikolayev #:
It looks like the "invention" of ensembling from MO is coming.

Wrong guess, there is no ensemble. The principles have been stated many times already.))
1...636465666768697071727374757677...86
New comment