in principle the article is not bad, everything is clear, clear, so to say on the shelves ..... let's say there is a place to start, not to search other forums, but it is necessary to continue, part 2,3.....
you need an example of working with multilayer NS, because the pictures show multilayer NS, but the example is only for a network with one neuron - either rename the article or write part 2.
good luck!

- www.mql5.com
the article is good, introductions are revealed. but I didn't quite understand - how the "training" of neurons takes place, and how often it is nano.
the article is good, the introductions are covered. but I didn't quite understand how neurons are "trained" and how often it is done.
- www.mql5.com

- www.mql5.com
I will support you in turn!!! and we will post one after another )))))).
I was dealing with neural networks for a long time - I couldn't understand how they work, I understood the mathematical apparatus, there are plenty of examples, but I couldn't understand why it is necessary to train them and what it gives for a long time because there were only examples of training for a specific task (pattern recognition, trading) in the Internet, and I didn't meet any elementary (primitive, everyday) examples, until I met an example of training sin(x)...... on some forum.
Then I wrote for myself a script that taught NS multiplication table online and printed the results to the screen using Comment(), and .... oh wonder... I had tears in my eyes from happiness, I saw how NS learns, really learns, not dry formulas to get the desired result. It would be nice to show an example in the article as an example 2x2=4,1 2x3=5,9 ..... 9x9= 80,2 , but 11x2 = 10,3 because the multiplication table for 11 NS has not studied, here is howtoktotak ))))))
ZY: about me: I studied, studied, served, worked, married, live happily, and died on the same day and I'm not going to die yet - there are still unfinished business !!!!
Good article for entry level, would like a follow up too.
Then I wrote for myself a script that taught the NS multiplication table online and printed the results to the screen using Comment(), and ... oh wonder... I had tears in my eyes from happiness, I saw how NS learns, really learns, not dry formulas to get the desired result. It would be nice to show an example in the article as an example 2x2 =4,1 2x3=5,9 ..... ..... 10x10= 10.2 , but 11x2 = 10.3 since the multiplication table for 11 NS has not studied, wotkaktotak ))))))
That's a good way of putting the question.
Although the multiplication table has little to do with trading, I still wonder if the network has been able to teach the multiplication table?
Perhaps by answering the question, why can't the network learn the multiplication table, there will be an understanding of how to teach the network to trade.
Good question statement.
1.Although the multiplication table has little to do with trading, I still wonder, has it been possible to teach the network the multiplication table?
2.Perhaps by answering the question, why can't the network learn the multiplication table?
3.There will be an understanding of how to teach the network to trade.
1.yes, NS was taught the multiplication table from 1x1 to 9x9 inclusive, the video seems to show the results to the 2nd decimal place.
2. the network has not learnt the multiplication table more than 9x9 , which is not surprising - no one taught it, any coincidences with the correct results are random.
3. There is an understanding, but there was hope for the developers of "New Neural" - Open Source project of neural network engine for MetaTrader 5 platform, but as they say "and there is still nothing" (Krylov). It is a labour-intensive task to develop a project of multilayer NS - it is easier to add .dll , although, probably, on pure mql5 there will be a loss in training speed in comparison with ready-made implementations of NS packages/.dll .
SZY: You can ask in the topic with MT5 wishes for an inbuilt implementation of multilayer NS, but as it happens with developers - they answer no, it is not provided, and in half a year they announce such an update, two times I already "got it": I wished for operator overloading and canvas for drawing on the chart :).
ZYZY: for me for experiments with NS such import is enough:
#import "fannmql.dll" void fannmql_build(string NN_config); // creates a FANN neural network, properties of neuron layers in NN_config // you can use delimiters: ', . ; -' or space void fannmql_unbuild(); // deletes the FANN neural network double fannmql_train(double &input[],double &output[]); // training FANN neural network, returns RMS value void fannmql_run (double &input[],double &output[]); // calculate FANN of the neural network void fannmql_savetofile (string File_name); // saving FANN of the neural network to a file void fannmql_loadfromfile(string File_name); // loading FANN of neural network from file #import string s = "8-10-10-2"; // NS with 8 inputs, 2 hidden layers of 10 neurons in each layer and 2 outputs double inp[8],out[2]; //________________________________________________ int init(){ fannmql_build(s); ..............but, apparently, to build such functions into MT5 standard functions is a grand_and_hard_to_realise_problem;)))))
Perhaps by answering the question, why can't the network learn the multiplication table?, it will be possible to understand how to teach the network to trade.
The most interesting thing is that the network cannot be taught the multiplication table (in the broad sense). That is, of course, you can teach multiplication from 1*1 to 9*9, and then ... Repin's painting - "Fig you".
But the brain performs these steps without difficulty, why is that so? - This is where the treasure trove of wisdom lies, you have to dig here, but it's not as simple as it may seem - it borders on the field of artificial intelligence, and that's a whole other story....
The most interesting thing is that the network cannot be taught the multiplication table (in the broad sense). That is, of course, you can teach multiplication from 1*1 to 9*9, and then... Repin's painting - "Fig you".
But the brain performs these steps without difficulty, why is that so? - This is the treasure trove of wisdom, we need to dig here, but it is not as simple as it may seem - it borders on the field of artificial intelligence, and this is a different story....
I don't agree about figwam.
Isn't neural network from the field of artificial intelligence?
If there is a desire to continue this topic, I suggest to go here "New Neural" - Open Source project of neural network engine for MetaTrader 5 platform.
Or via private message, I think it will be redundant here.
The most interesting thing is that the network cannot be taught the multiplication table (in the broad sense). That is, of course, you can teach multiplication from 1*1 to 9*9, and then... Repin's painting - "Fig you".
But the brain performs these steps without difficulty, why is that so? - This is the treasure trove of wisdom, we need to dig here, but it is not as simple as it may seem - it borders on the field of artificial intelligence, and this is a different story ...
Yura Reshetov even wrote that it is possible to teach the NS the multiplication table, where 2*2 = 5, and all other variations will be arithmetically correct. That's the essence of fitting, by the way. And about why the NS will start lying if you teach it 1-9 and feed it 10-20, there is also an answer. The point is that the network can work correctly only and only if the out-of-sample data lie in the same interval as on the training set. Roughly speaking, our inputs will be in the range [1;9]. And we will then give the NS data in a different range, which will cause the input data to be non-stationary, and non-stationarity will give false answers. Try to train the network on samples in the range [1;20], and then give it other samples in the same range, and, oh wonder, it will really work. You can test it.
Here I have tested it. Works great out of sample )
NS was taught the multiplication table 1-20. Training and stopping of training was done on randomly selected examples (1/3 and 1/3 of the sample size). Validation - the remaining 1/3 of the sample size, but most importantly from the same range. The results on validation are as good as on training samples. The error is within +-0.1.
The point is that NS learns any function, as you know, and does it successfully, the main thing is that the range of out-of-sample data should be inside the training range.

- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
New article Neural Networks: From Theory to Practice is published:
Nowadays, every trader must have heard of neural networks and knows how cool it is to use them. The majority believes that those who can deal with neural networks are some kind of superhuman. In this article, I will try to explain to you the neural network architecture, describe its applications and show examples of practical use.
The Concept of Neural Networks
The artificial neural networks are one of the areas in artificial intelligence research that is based on the attempts to simulate the human nervous system in its ability to learn and adapt which should allow us to build a very rough simulation of the human brain operation.
Curiously enough, artificial neural networks are made up of artificial neurons.
Author: Дмитрий