Neural networks. Questions from the experts. - page 17

New comment
 
VladislavVG:

SVM .... http://www.csie.ntu.edu.tw/~cjlin/libsvm/


Thanks. I will read it...

If you don't mind telling me, comparing to FANN, what are the advantages, where are the weaknesses, what are the difficulties?

................................

A question for everyone:

What do you use? How is it implemented?

 
lasso:

Thank you. I'll keep reading...

If you don't mind telling me, comparing with FANN what are the advantages?, where are the weaknesses?, what are the difficulties?

................................


https://ru.wikipedia.org/wiki/%D0%9C%D0%B5%D1%82%D0%BE%D0%B4_%D0%BE%D0%BF%D0%BE%D1%80%D0%BD%D1%8B%D1%85_%D0%B2%D0%B5%D0%BA%D1%82%D0%BE%D1%80%D0%BE%D0%B2

Advantages and disadvantages of SVM:

    it is the fastest method of finding solver functions;
  • the method reduces to solving a quadratic programming problem in a convex domain that always has a single solution;
  • the method finds a separating band of maximum width, which further allows for more confident classification;
  • the method is sensitive to noise and data standardization;
  • there is no general approach to automatic kernel selection (and construction of the conjugate subspace in general) in case of linear non-separable classes
    • .

This is pdf... tht extension remove.
Files:
svm.pdf.txt  288 kb
 
lasso:

Here's what my Stats 6 generates ))

What's the joy?

The joy is just to see in the debugger the signal from input to output.

The joy is in visual creation and training of the network. It is convenient in data preparation.

The disadvantage - I will have to hack a file to develop it into a dll or to transfer it to MQL

lasso:

Or I'm generating the wrong data in the wrong place?

Or am I generating in the wrong place?


Currently I'm writing the network in MQL.

It is convenient - I immediately get an Expert Advisor, for which I am writing the network.

The disadvantage is low speed of MQL4 and limited possibilities of data visualization. In this regard, MQL5 is much more preferable.

For the study of neural networks I recommend using the specialized programs that allow working with them (you are already familiar with Statistica), and for the implementation of complex non-standard projects it is still better to work in the environment, where the direct work with networks will be needed - MetaTrader.

And when the idea has been tested and worked out, it is never too late to transfer the grid into a dll, simplifying and speeding up the work with the grid.

 

Hello Gentlemen,

Can you tell me what the problem could be?

I am building the NS in NeuroSolutions software. It has an option to compile the NS project into a dll. In this process I have to choose a programming language which the file will be compiled in. Following the advice from the article(https://www.mql5.com/ru/articles/236) I choose visual c++ 6 and installed visual c++ 6 on my computer. But it fails, .dll creation failed. I read the same article about possible reasons for the failure. All I got was that visual C++ 6 may be configured incorrectly. Has anyone encountered such a problem with NeuroSolutions? What is the first thing to try and change?

 

joo:
Statistica генерит С-шный исходник консольного приложения обученной нейронной сети (если скомпилировать такой исходник получится exe-исполняемый файл).

......

Disadvantage - I'll have to make it into a dll or transfer it to MQL.

Now I've looked at my past experiments with Statista... I've used dlls derived from it everywhere, I don't remember any broken bits, at most I just added them to the library header (something like this pops up in my memory). If the redesign was serious - would remember for sure, I'm not God knows what kind of C-programmer) And note all worked as it should))) Although, I had version 8, maybe there are differences in this regard...
 
Figar0:
Now I've looked at my past experiments with Statista... I use dlls derived from its code everywhere, I don't remember having to work with a file, at most I added to the header of the library (something like this pops up in my memory). If the redesign was serious - would remember for sure, I'm not God knows what kind of C-programmer) And note all worked as it should))) Although, I had an 8 version, maybe there are differences in this regard ...
But you still have to poke around in the code, and every time after training.
 
Figar0:
Now I looked at my past experiments with Statistica... I've used dlls derived from its code everywhere, I don't remember having to work with a file, at most adding to the header of the library (something like this pops up in my memory). If the redesign was serious - would remember for sure, I'm not God knows what kind of C-programmer) And note all worked as it should))) Although, I had version 8, maybe there are differences in this regard...

I'm having some vague doubts.....

Here's the latest info from the website on the 8, I understand.

-------------------------

Source Code Generator

The source code generator is an add-on product that allows users to easily create their own applications based on the STATISTICA Neural Networks system. This add-on product creates system source code of the neural network model (as a C, C++, C# file) that can be separately compiled and integrated into your program for free distribution. This product is designed specifically for corporate system developers as well as for those users who need to convert highly optimized procedures created in STATISTICA Neural Networks into external applications to solve complex analytical problems. (It should be noted that to get permission, users should inform StatSoft Russia about the distribution of programs using the generated code).

----------------------------------

Can you attach the inline bible?

Especially if it's generated by StatSoft, it's not a secret.

 
lasso:

Can you attach the inclusive biblio?

Especially if generated by Statistics is not a secret.


The source code is just not left, dlls are lying around, but what is the use of them without source code and without training? I myself do not remember what's in them) Now I'm just about to put 8 again, there are developments in which it can help. I'll show you something, of course, but it's easier to try it yourself.
 
Figar0:

I don't have any source code, I have some dlls lying around, but what's the use of them without source code or training? I can't even remember what's in them myself) Now I'm just about to put 8 back on, there are developments in which it can help. I'll show you something, of course, but it's easier to try it yourself.

I've never even asked for sources, only the dll and header.

You wrote it yourself...


Figar0:
Now I looked at my past experiments with Statistica... I've used dlls derived from its code everywhere, I don't remember having to work with a file, at most I added a header to the library (something like that pops up in my memory). If the redesign was serious - would remember for sure, I'm not God knows what kind of C-programmer) And note all worked as it should))) Although, I had version 8, maybe there are differences in this regard...
 

Whatever networks you use, there is always only one question: how do you NOT fit a network into a past market?

1...10111213141516171819202122232425
New comment
Reason: