neural network and inputs - page 38

New comment
 

Good afternoon.

Not only that. I use ada, randomForest ( in various variants), C50 and others. But the best results, of course, come from ada and rfNear() from CORElearn. The highly praised svm has not shown any advantages (except that it takes a very long time to learn). Without belittling its importance, maybe I couldn't cook it right?

The sequence of work is as follows.Check the input data for significance, determine the optimal parameters for each model. Calculation by three different models with calibration, decision by simple voting. With initial sample of 1000 bars model stably works with 250 bars without retraining. Accuracy controls and estimates.

There are packages that estimate the influence of preprocessing methods on the result and select the best one in addition to the evaluation of input data using principal components methods. I haven't mastered them yet, as I don't have enough time. But I will continue studying them.

Good luck.

 
grell:

Which input signals do you think carry useful information?:)

See PDF in the attached archive: p. 17 Reduction of insignificant predictors ...


There's a simple example in there, which clearly shows that the method correctly downsamples only those inputs which are not explanatory variables in the training sample.

 
HMM... Interesting, downloaded the project from the link in the book but don't know how to run it. Can you give me a hint?
 
And when opened via java. It says no Jproject class.... Like this....
 

Here's a window like this....

 
nikelodeon:
HMM... Interesting, downloaded the project from the link in the book, But how to run it I don't know. Can you tell me???

Is it possible that the version of Java on your computer is out of date? I compiled the project under version 1.8.0_25

Download the latest version of Java2SE (JRE or JDK) from: http: //java.com/ru/

 

Great, everything is running, but it is not clear in what form the data should be submitted.... It is clear that .csv Well, how they should be arranged, etc.

If you can send me a data file with an example....

 
nikelodeon:

Great, everything is up and running, but it's not clear what form the data should be in.... It's clear that .csv Well, how should they be arranged, etc.

If you can send the data file with the example to....

see. Forecasting bankruptcy.

There is a CSV file attached with an example for classification.

In short:

The cell separator is a semicolon.

Separator of integer part from fractional part for numbers: either dot or comma (all commas are automatically replaced by full stops).

The first column - examples identifiers (in textual form)

Last column - values of dependent variable: 1 - belonging to a given class, 0 - belonging to an undefined class (any class other than the given one)

First line - factor identifiers (in textual form)

Second line - notes on factor identifiers, e.g., units of measurement (as text)

The remaining data below the second row, to the right of the first column, and to the left of the last column are numeric values of factors (non-numeric are not processed).

 
Of course my formula was simpler, and I didn't have much data, but how do I interpret it? I don't understand :-(.
 

I see. The first transformation is normalization of the input data, after normalization I substitute the data into the formula and get the RESULT of WHAT? I mean, I put the normalized data into the formula and get the curve of the indicator according to the formula??? Why should I put a sign >0???

it's a bit unclear :-(

1...3132333435363738394041
New comment
Reason: