"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 67

[михаил потапыч]

TheXpert:

Tell me what you're going to do. It's all done for the programmer. I.e. API.

And for the user there is nothing, I usually automate, with minimal possibility of intervention on the part of unauthorized hands. In other words, no user interface.

Told you so ((

Don't bother.

[TheXpert]

Mischek:

Told you so ((

Then neuro-packages, pick them. Try to look from the opposite side of yours, you can't get a "pish" button, only for a specific task.

You need to prepare the data, feed it correctly, train it correctly, and evaluate the results correctly.

[Mykola Demko]

Actually, this is in the theme "Interesting ...", but the subject is closer here.

Evolution and artificial life

and there's life here http://www.math.com/students/wonders/life/life.html

[Mykola Demko]

Gentlemen, who is familiar with the learning algorithms listed below,

List the variants of the move in training.

For example in the case of backpropagation, first there is a forward move of calculations, then a backward move of error propagation.

What other variants of the move are there in numerous learning algorithms?

SZ This is needed in order to put the necessary virtual functions in the engine.

ZZZY below is a table, who knows what write what option (s) move used in these algorithms.

[Mykola Demko]

Paradigm	Learning rule	Architecture	Learning algorithm	Task
With the teacher	Error correction	Single and multilayer perceptron	Perceptron Learning Algorithms Reverse Propagation Adaline and Madaline	Pattern classification Function approximation Prediction, control
	Boltzmann	Recurrence	Boltzmann Learning Algorithm	Pattern Classification
	Hebb	Multilayer Direct Propagation	Linear Discriminant Analysis	Data Analysis Pattern Classification
	Competition	Competition	Vector Quantization	In-class categorization Data compression
		ART Network	ARTMap	Classification of Images
Without a teacher	Error correction	Multilayer direct propagation	Sammon projection	In-class categorization Data analysis
	Hebb	Direct propagation or competition	Principal component analysis	Data analysis Data compression
		Hopfield network	Learning associative memory	Associative memory
	Competition	Competition	Vector Quantization	Categorization Data compression
		Kohonen SOM	Kohonen SOM	Categorization Data analysis
		ART networks	ART1, ART2	Categorization
Mixed	Error correction and competition	RBF network	RBF learning algorithm	Pattern classification Feature approximation Prediction, control

[Mykola Demko]

Okay, the general silence so I understand from the complexity of the question.

Let me rephrase the question:

For which learning algorithm, the backtracking of layers is not appropriate?

[Serj]

Urain:

Okay, the general silence so I understand from the complexity of the question.

Let me rephrase the question:

For which learning algorithm, the backtracking of layers is not appropriate?

Rather, "For which learning algorithm, backtracking of layers is not necessary?"

I know one thing for sure, genetic algorithm doesn't need it.

In other cases, I could be wrong, it is.

[Mykola Demko]

her.human:

Most likely "For which learning algorithm, the backtracking of layers is not necessary?".

I know one thing for sure, the genetic algorithm doesn't need it.

In other cases, I could be wrong, it is.

The point is, set for the learning algorithm (to adjust weights) only the reverse stroke of the enumeration of layers, if the learning itself is not needed then you can call a straight move of the calculation of the grid, if you need it, then call and reverse.

Just me a doubt, maybe some algorithm needs the forward stroke of grid computation and the forward stroke of weights adjustment?

I really don't know such algorithms, but I can't know everything.

[Alexey]

Good afternoon, not really on topic.

I myself set a task before me. It is necessary to choose an adaptive time window of the current moment, and not to set it in the parameters, such as 10 bars. Then run through the history in depth to determine which cluster the selected window belongs to. Can neural networks handle it, or is it easier to do something else? If you don't mind, please send me a book on grids at the sausage level.

[yu-sha]

"Reverse" is needed to find partial derivatives of a fitness function by adaptive arguments, i.e. any gradient methods (e.g. BackProp in any modification) require "reverse"

Other methods do not require