I searched in this forum but I could not found a topic talking about it.
I´m programming an EA that has a neural network. I structure the code to function like a network and developed a specific architecture according that I experienced in the market. I´m using as a model the article made by Yury Reshetov (https://www.mql5.com/en/articles/1447)
It has four layers with 16 source neurons, 10 hidden neurons and 1 exit neurons implemented in the code.
According with the article, we could train the neurons using the genetic algorithm include in the tester. We could set to optimize the weights and the tester will randomize the weights until found the best set of parameters.
But, searching more about trainning methods, I verified that exist other methods with better accuracy and faster than GA. One example is standard backpropagation.
The main question is, we could implement or add another algorithm for optimization/trainning in the Mt4? The function of this algorithm is only adjust the weights of the neural network implemented in mql, reaching the best set of weights.
I read some things about using matlab or FANN2mql package but it is very complex and, thinking about a commercial EA, all buyers platform will need has installed in the metatrader packages to run the neural network.
I think it´s very simple to create a neural network using mql. We could easily create whatever neural network architecture in the code, but we can use only one algorithm to train the network.
If someone could explain an easier solution to implement another algorithm to optimize the network, I will be very grateful.