Finding a set of indicators to feed into the neural network inputs. Discussion. A tool for evaluating the results. - page 7
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
to iliarr
Try, for the sake of interest, to use the number of trades, or the relative drawdown as a fitness function. You may not trade at all if you use drawdown as a fitness function.
It may be that the net will not trade at all if you use drawdown as a fitness function. :))
to iliarr
Try, for the sake of interest, to use the number of trades, or the relative drawdown as a fitness function. You may not trade at all if you use drawdown as a fitness function.
It may be that the net will not trade at all if you use drawdown as a fitness function. :))
I thought about it, but decided to postpone it for a while...
If the target function is only the number of trades or only the drawdown, it will be of little use, because the network will either learn to enter/exit the market often and aimlessly or learn to avoid drawdowns....
I have to optimize both profit and the number of trades and drawdown... As I remember JGAP allows to have target function with several outputs... my current priorities are: to solve input data and refine recurrence neuronet.
as i see, at the moment nobody is interested in searching and testing input data with the method i suggested...
Come to a common denominator on testing? ;-). Here is a suggestion about this. If a network is trained without a teacher to make hypothetically unlimited profit, it should be kept in mind that the input data still imposes a limit from above on the size of the profit. It is possible to estimate the amount that cannot be exceeded by the selected learning period (with a constant lot, by the selected strategy). Thus, we can calculate the learning ratio of the grid for this period as a ratio of the theoretical maximum possible profit to the profit that the grid gives. Then similar estimations are performed for the validation period and the ratios are compared.
As it's been pointed out here, without such a check, it's worthless, imho.
If a network is trained without a teacher to make hypothetically unlimited profits
This is called overtraining. we have already raised this issue.
Question for amateurs looking for neural network inputs :)
Has anyone dabbled with principal component analysis (aka "principial component analysis" or "pca")?
Question for amateurs looking for neural network inputs :)
No one dabbled with principal component method (aka "principial component analysis" or "pca")?
How are you going to apply it?!
Question for amateurs looking for neural network inputs :)
No one dabbled with principal component method (aka "principial component analysis" or "pca")?
I developed a system based on the GHA algorithm. It works very well if there is noise. You can do it via DFT or you can do it via principal component analysis. Please refrain from using the word "amateurs" without the prefix "dear" :)lea писал(а) >>
Has anyone dabbled with principal component analysis (aka "principial component analysis" or "pca")?
I have, but in a completely different field of application. By the way, I have never been able to make non-linear PCA work. And the linear one, I think, is weak.
I developed a system based on the GHA algorithm. It works fine if there is noise. You can do it via DFT, or you can do it via principal component analysis.
What was all this calculated in? MathCad/MathLab?
Please refrain from using the word "amateurs" without the prefix "respected" :)
OK :)
I dabble, but in a very different field of application. By the way, I failed to make non-linear PCA work. The linear one, imho, is a bit weak.
So far I hope to make do with the linear one.
How are you going to use it?
By its intended purpose - to select a set of variables, which will be correlated more loosely than the original ones.