Discussion of article "Deep Neural Networks (Part I). Preparing Data"

 

New article Deep Neural Networks (Part I). Preparing Data has been published:

This series of articles continues exploring deep neural networks (DNN), which are used in many application areas including trading. Here new dimensions of this theme will be explored along with testing of new methods and ideas using practical experiments. The first article of the series is dedicated to preparing data for DNN.

Ris5

Fig.5. Oscillators of the third group

Author: Vladimir Perervenko

 
MetaQuotes Software Corp.:

The article Deep neural networks (Part I) has been published . Data preparation:

Author: Vladimir Perervenko

Vladimir. As always, thanks for the quality article. I will definitely work through it.

 

Digital filters + diplerning is an interesting combination, but the digital filters themselves should be selected by the generator on the machine in the course of training, otherwise it is pointless, IMHO

 
RTrade:

Vladimir. As always, thanks for the quality article. I will definitely work through it.

Part_2 (development and selection of predictors) should be out next week. I will need the results of the first part there.

Good luck

 
Vladimir Perervenko:

Part_2 (predictor development and selection) should be out next week. You will need the results of part 1 there.

Good luck

Vladimir.

Regarding the last series of articles. I ran robots on a rented server on Amazon. Out of 40 models 5 were in plus.

This is a good result for me. Quotes data, which are broadcasted by the broker, differ from the archive data. Therefore, the models did not work on the data they learnt on.

These things are quite transferable to more understandable markets. If we managed to show a positive result on forex with its volatility and divorces.

 

very interesting article

I've always wondered why they always choose exotic programming languages for neural networks (more often php, here R).

What's wrong with C#, which is used by 3/4 of the world?

 
gedd:

very interesting article

I've always wondered why they always choose exotic programming languages for neural networks (more often php, here R).

What's wrong with C#, which is used by 3/4 of the world?

The R language is not exotic at all. It is de facto and de jure a standard language for statistical calculations. C# and R are different levels of abstraction, the first is the level of implementation, the second is the level of models.

If you google, you will be surprised by R's place in the ranking of languages. And as for 3/ 4 - you are overstating it. It all depends on the tasks to be solved.

Good luck

 
gedd:

very interesting article

I've always wondered why they always choose exotic programming languages for neural networks (more often php, here R).

What's wrong with C#, which is used by 3/4 of the world?

It is purely as a framework for statanalysis and so on. All basic libs are written in C++ with neural networks, you can easily find the site of a particular lib and connect it as a dll to MT5, for example.

and do all the necessary research on R first

 
Maxim Dmitrievsky:

It is purely as a framework for statanalysis and so on. All basic libs are written in C++ with neural networks, you can easily find the site of a particular lib and connect it as a dll to MT5, for example.

and do all the necessary research on R first.

I don't understand the highlighted part. Any example?
 
Vladimir Perervenko:
I don't understand the highlighted part. Any example?

Well http://mxnet.io/ is the same, you can build as a dll, and some packages come with ready-made dlls.

or what example how to use third-party libraries in MT5?

p.s. there is a ready-made one, you don't need to build anything http://mxnet.io/get_started/windows_setup.html.

 
Maxim Dmitrievsky:

Well http://mxnet.io/ is the same, you can build it as a dll, and some packages come with ready dlls.

or what is an example of how to use third-party libraries in MT5?

Yes, there are some models that require the creation of a dll, but they are very few and their dlls do not need to be connected to MT. The DLL is simply installed in your system. The latter is often not easy.

Good luck