Indicators: BPNN MQL Predictor Demo with library

[Automated-Trading]

BPNN MQL Predictor Demo with library:

This is a demo indicator with BPNN neural network library ported from C++ to MQL.

Author: Stanislav Korotky

[chcharitos]

Excellent piece of work! If I understand correctly the neural network takes into account past values of timeseries to predict the next right?

Could you please tell me how to put rsi indicator into the code along with open high low close values so the neural network predicts the next value? Is it possible?

[Stanislav Korotky]

chcharitos:

Excellent piece of work! If I understand correctly the neural network takes into account past values of timeseries to predict the next right?

Could you please tell me how to put rsi indicator into the code along with open high low close values so the neural network predicts the next value? Is it possible?

Yes you're right, but, actually, this is not an example on how prediction of price timeseries should be made using BP NN - this is an example of the library usage accompanied with the library itself. For proper prediction many other stages (such as statistical analysis, normalization, folding into higher dimesions, taking other factors as inputs (say, day of week, hour of day, currency-related indices, etc.)) should be performed, they are left out considerations here. Probably you should read documentation of the original indicator and other staff gpwr published.

If you want it quick and dirty, please, find the line in the indicator's source code where the handle for iMA is obtained. Then replace iMA instatiation with iRSI.

[chcharitos]

Stanislav Korotky:

Yes you're right, but, actually, this is not an example on how prediction of price timeseries should be made using BP NN - this is an example of the library usage accompanied with the library itself. For proper prediction many other stages (such as statistical analysis, normalization, folding into higher dimesions, taking other factors as inputs (say, day of week, hour of day, currency-related indices, etc.)) should be performed, they are left out considerations here. Probably you should read documentation of the original indicator and other staff gpwr published.

If you want it quick and dirty, please, find the line in the indicator's source code where the handle for iMA is obtained. Then replace iMA instatiation with iRSI.

Thank you very much!

[farshad khoshlahn]

Hello , thanks for your code.I am beginner and i dont know how use iRSI in your code.

Also i am from iran and Iran stock work only Saturday till Wednesday on 8:30am till 12:30 pm.if possible please change your code with iRSI for Predicate market. 

with best regards 

Farshad

[Cristian Eriksson]

Hi, thanks a lot for this work and publication.

In the linked original publication of this library. It's stated that it contains memory leaks, has that been fixed in this version?

[Stanislav Korotky]

Cristian Eriksson:

Hi, thanks a lot for this work and publication.

In the linked original publication of this library. It's stated that it contains memory leaks, has that been fixed in this version?

That leak was specific to C/C++. This version has all multidimensional arrays re-implemeted in MQL.

[Alik Dmitriev]

How to save and load the network?

[Stanislav Korotky]

Alik Dmitriev #:
How to save and load the network?

You may consider storing/loading extInitWt[] array of weights along with number of layers and their sizes. Also you may preserve activation functions used.

You can check this article to look how it's implemented with the same NN re-implemented on the matrices. The article contains the section Predictive Indicator (with similar indicator) and the section Storing NNs in Files.

Please, note that there is a bugfix for the article. I've asked moderators to update the source code, but it didn't happen so far.