Question on neural network programming

New comment
 
Good time everyone!
I have a question about programming .
I would like to know what data is fed to the inputs of a neural network to predict the closing price of a bar.
If possible, give an example of code for calculations.
Who, what formulas do they use for calculations?
There are a couple of neural network-based EAs, but they earn slightly more than they lose. Well, this is not serious.
 
Bar closing price forecasting is the worst option. It is the nastiest as it is the most informative. It is better served as an input. And this is not just my imho. See my last comment in the 'Graphical Analysis. Are there any developments?".

What kind of network do you use (architecture)? How do you prepare data? I don't have code, but have a little experience with interpolating nerfs. The experience is negative. If you want, write to my email, it's in my profile.
 
Read the book by Ezhov and Shumsky "Neurocomputing and its application in economics and business". It describes very well how to prepare input data correctly :) Good luck!
 

For example, when I worked with neural networks, I used the following scheme:

  1. I'd select a signal from the current bar of some length
  2. decompose it into single sinusoidal pulses (Fourier transform, only slightly more complicated),
  3. and then let a neural network descend on those impulses to make a forecast of a new impulse.
  4. After getting its characteristics, I performed the reverse procedure: built a superposition of impulses and obtained a forecast of a price series
 

Grasn, have you tried forecasting swings with NS? It's still a more natural organisation of data than bars. This idea has been keeping me awake lately...

 
Mathemat:

Grasn, have you tried forecasting swings with NS? It's still a more natural organisation of data than bars. This idea has been keeping me awake lately...

I should clarify, I have not predicted exactly the bars. I chose the initial signal of the following form: (H+L)/2. I have made two modifications of the system:

  1. The first model was based on Eliot's theory. It is a well-known fact that if you select harmonics, you can build any pattern of the wave theory. The same is with impulses. Having obtained the forecast impulse (let me remind you, my NS made impulses forecasts, not prices) I performed analysis based on my knowledge base (i.e. different impulse structures and patterns correlated to them) and selected the most probable movement model
  2. The second model gave more practical results. Having got a forecasted signal (I collected a superposition of impulses including the forecasted one) I calculated some "average" value, which I used only under certain conditions as a level (possible profit), otherwise the forecast was ignored.

PS: I would like to add that I finally parted with NS about five years ago. Reliable to use them is an illusion. It may take more time to draw such conclusions. No NS will be able to consistently predict the price range with an acceptable probability. And they work great of course, but only in individual "academic" cases. And you'll sooner recognize a launcher in a picture of your garden plot than distinguish the third wave from the fifth. But this is, in a way, a digression. :о)


PS2 Reread my post and realized that I have not answered your main question. The answer is, I am sleeping peacefully and I do not dream of successfully predicting NS anymore.

 

Thanks, grasn. I too stopped dreaming after my amateur experiments (a couple of years ago), but apparently I haven't worked that cycle out yet - especially since I haven't even taken up the qualifying NS...

 
Mathemat:

Thanks, grasn. I too stopped dreaming after my amateur experiments (a couple of years ago), but apparently I haven't worked that cycle out yet - especially since I haven't even taken on qualifying NS...

Yeah, you're welcome. Don't mind my pessimism about NS. Everyone has to go their own way. By the way, long time researches of signal structure helped me a lot to develop the model which I developed on the materials of a friendly forum(https://www.mql5.com/ru/forum/50458). It so happened that the ideas stated by Vladislav and many other participants of discussion (I don't mean Alex) are very well laid on my own experience and understanding of processes.

PS: By the way, I recommend MineSet for research (if any pattern needs to be found), developed by SGI and sold here: http://www.purpleinsight.com/ when SGI collapsed. There is a necessary set of Data Mining tools including classification, as well as excellent visualization possibilities (after all, SGI created it, and no one has invented better than the eye).

 
Mathemat:
Bar closing price forecasting is the worst option. It is the nastiest as it is the most informative. It is better served as an input. And this is not just my imho. See my last comment in the 'Graphical Analysis. Are there any developments?".

What kind of network do you use (architecture)? How do you prepare data? I don't have code, but have a little experience with interpolating nerfs. The experience is negative. If you want, write to me, it's in my profile.
I use multilayer perseptron.
I am inputting a closing price of 5 bars, and then on change of a signal on output (>0 or <0) commands Buy or Sell are executed.
 
Yes, dob-zorge, that's what you should be feeding it into, not predicting it.
 
plan:
Read the book by Ezhov and Shumsky "Neurocomputing and its application in economics and business". It describes very well how to prepare input data correctly :) Good luck!
Thanks for the tip!
I will study, if everything will work out, I will lay out a code of the Expert Advisor for testing.
1234567
New comment
Reason: