Finding a set of indicators to feed into the neural network inputs. Discussion. A tool for evaluating the results. - page 9

New comment
 
lea >> :

I'm not really looking for a price series wrap. There is already a set of indicators (i.e. a transformed price series), so the dimensionality of this set needs to be reduced.


I recommend that for this purpose you should also consider the possibilities of co-organisation maps. The dimensionality is better reduced with them. They will also highlight the array of states. There is also such a variant of analysis as analysis of independent components. It is much more promising, but so far I have not been able to grasp it to the full extent. If you are interested in the analysis of independent components - let me know.
 
joo >> :

Here are a couple of books on optimisation. Just downloaded, still hot.

........ can't seem to attach. I got it from http://torrents.ru.


You could have clarified the direction a little bit.
 
IlyaA >> :
You could have clarified the direction a bit.

On demand "optimisation":


http://torrents.ru/forum/viewtopic.php?t=1591908e

http://torrents.ru/forum/viewtopic.php?t=2139370e

http://torrents.ru/forum/viewtopic.php?t=1327023e

http://torrents.ru/forum/viewtopic.php?t=711214e

http://torrents.ru/forum/viewtopic.php?t=2346898e

http://torrents.ru/forum/viewtopic.php?t=2123107e

 

Thanks joo. Intriligator was my board book. :) Check it out, I mean, study the material.

 

This is what a mathematician friend wrote to me when asked how to get rid of input correlation:

"The principal component method seems to be it. Here is a more or less accessible description: http://www.statsoft.ru/home/ textbook/modules/stfacan.html. How effective it is in your case, I don't know. But linear correlation should eliminate it well. "

I'll look into it.

 

PCA is data transformation and dimensionality reduction, i.e. after data transformation we find the best new input data (according to the criterion).

If you need to select non-correlated data from the initial data, then multivariate regression rules. For example, if you have a set of indicators, you can carefully run a multivariate regression in a statistical program and find the right set of indicators.

 
IlyaA писал(а) >>

I also recommend you to look into the possibilities of soma-organization maps for this purpose. Dimensionality is better reduced with them. They will also select the array of states. There is also such variant of analysis as analysis of independent components. It is much more promising, but so far I have not been able to grasp it to the full extent. If you are interested in the analysis of independent components - let me know.

Self-organisation maps will be considered if the PCA fails.

I have heard about independent component analysis, but have not understood what it is in detail.

Now I plan to implement PCA based on my own matrix library.

IlyaA wrote >>

The calculation was done in excel.

Tough :)

 
lea >> :
IlyaA wrote(a) >>

the calculation was done in excel.

Tough :)

That's another thing... I'm catching myself thinking that I've started programming strength calculations in my head in MQL5.... :)

 
TheXpert >> :

So I must have used it incorrectly, or prepared it.

I myself have worked with picture compression. Sometimes the error is zero, sometimes not, depending on the degree of compression (number of principal components) and the informativeness of the inputs.

Try with simple examples.

Works great on attractors :) I haven't tried it on pictures. I think the question is the structure of the row. I haven't pre-processed it.

 
joo >> :

I agree with iliarr. It's gonna be a fit.

I suggested a variant without fitting. Fitting implies adjusting to a test set, and here the finished system is just tested on it. If you don't want to use it, don't use it, it's up to you.
12345678910
New comment