Probabilistic neural networks, packages and algorithms for MT4 - page 19

[[Deleted]]

I have tried working with the topology by changing the degree of signal smoothing or changing the input array qualitatively - the results are terrible. In probabilistic networks the untrained eye immediately notices several methodological contradictions encountered in network development - one of them is that the range of the test period is proportional to the non-linearity of the network. That is, it is not clear how to optimise the network.

[[Deleted]]

what can I say... here's the pornography))

[[Deleted]]

there she is

[[Deleted]]

xproit:
here it is

A quick glance. You're right - pornography.)

Why such a huge garden of completely uncoordinated inputs? What's the point of putting absolute values and their differences on the input at the same time? Nets get "drunk" from this...

[[Deleted]]

The thing is, I prepare a data file in MT4, putting everything I have into it, and in NEUROSHELL 2 I directly select, combine, etc.

[[Deleted]]

xproit:

The thing is, I prepare a data file in MT4 by attaching everything I have, and in NEUROSHELL 2 I directly select, combine, etc.

And can you show me the input data file for NS2?

And what is the criterion for the classification of the input set, ie why do you determine that this set of eg buy, this sell? Oops, saw it myself...

Z.U. By the way, I just dug up a relatively fresh NS2 a couple of days ago, just to experiment in it with PNN...

[[Deleted]]

I use nets in trading anyway. To confirm trade signals with predicted indicator values. This is probably the main advantage of this method over lagging and averaging. When predicting smoothed lines like BZL MACD(High,15,30) 3 points ahead even using lags of the indicator on entry I get correlation coefficient of 0.995 on average.

[[Deleted]]

In essence, probabilistic networks are less demanding in the task of determining the input array. The network in training uses an algorithm of individual corrections to the smoothing parameter of each input as well as the overall smoothing parameter. That is, during training the values of individual smoothing parameters are used as a tool for analyzing the sensitivity of the input The greater the parameter for a given input, the more important the input is for the model. That is, the network is not optimized by the input array. It is desirable to give it more inputs (candidates).

[[Deleted]]

Try to work with them I've been working with them for a while too. Here is an indicator for preparing a data file or rather a script

[[Deleted]]

xproit:
Try to work with them I have also been working with them for a while. Here is an indicator for preparing a data file or rather a script

I will have a look at it, thanks.